drivers/net/tun.c

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <[email protected]>
   4  *
   5  *  This program is free software; you can redistribute it and/or modify
   6  *  it under the terms of the GNU General Public License as published by
   7  *  the Free Software Foundation; either version 2 of the License, or
   8  *  (at your option) any later version.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
  16  */
  17
  18 /*
  19  *  Changes:
  20  *
  21  *  Mike Kershaw <[email protected]> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <[email protected]>
  25  *    Use eth_random_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <[email protected]>  2004/04/20
  28  *    Fixes in packet dropping, queue length setting and queue wakeup.
  29  *    Increased default tx queue length.
  30  *    Added ethtool API.
  31  *    Minor cleanups
  32  *
  33  *  Daniel Podlejski <[email protected]>
  34  *    Modifications for 2.3.99-pre5 kernel.
  35  */
  36
  37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  38
  39 #define DRV_NAME        "tun"
  40 #define DRV_VERSION     "1.6"
  41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  42 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <[email protected]>"
  43
  44 #include <linux/module.h>
  45 #include <linux/errno.h>
  46 #include <linux/kernel.h>
  47 #include <linux/sched/signal.h>
  48 #include <linux/major.h>
  49 #include <linux/slab.h>
  50 #include <linux/poll.h>
  51 #include <linux/fcntl.h>
  52 #include <linux/init.h>
  53 #include <linux/skbuff.h>
  54 #include <linux/netdevice.h>
  55 #include <linux/etherdevice.h>
  56 #include <linux/miscdevice.h>
  57 #include <linux/ethtool.h>
  58 #include <linux/rtnetlink.h>
  59 #include <linux/compat.h>
  60 #include <linux/if.h>
  61 #include <linux/if_arp.h>
  62 #include <linux/if_ether.h>
  63 #include <linux/if_tun.h>
  64 #include <linux/if_vlan.h>
  65 #include <linux/crc32.h>
  66 #include <linux/nsproxy.h>
  67 #include <linux/virtio_net.h>
  68 #include <linux/rcupdate.h>
  69 #include <net/net_namespace.h>
  70 #include <net/netns/generic.h>
  71 #include <net/rtnetlink.h>
  72 #include <net/sock.h>
  73 #include <net/xdp.h>
  74 #include <linux/seq_file.h>
  75 #include <linux/uio.h>
  76 #include <linux/skb_array.h>
  77 #include <linux/bpf.h>
  78 #include <linux/bpf_trace.h>
  79 #include <linux/mutex.h>
  80
  81 #include <linux/uaccess.h>
  82 #include <linux/proc_fs.h>
  83
  84 static void tun_default_link_ksettings(struct net_device *dev,
  85                                        struct ethtool_link_ksettings *cmd);
  86
  87 /* Uncomment to enable debugging */
  88 /* #define TUN_DEBUG 1 */
  89
  90 #ifdef TUN_DEBUG
  91 static int debug;
  92
  93 #define tun_debug(level, tun, fmt, args...)                     \
  94 do {                                                            \
  95         if (tun->debug)                                         \
  96                 netdev_printk(level, tun->dev, fmt, ##args);    \
  97 } while (0)
  98 #define DBG1(level, fmt, args...)                               \
  99 do {                                                            \
 100         if (debug == 2)                                         \
 101                 printk(level fmt, ##args);                      \
 102 } while (0)
 103 #else
 104 #define tun_debug(level, tun, fmt, args...)                     \
 105 do {                                                            \
 106         if (0)                                                  \
 107                 netdev_printk(level, tun->dev, fmt, ##args);    \
 108 } while (0)
 109 #define DBG1(level, fmt, args...)                               \
 110 do {                                                            \
 111         if (0)                                                  \
 112                 printk(level fmt, ##args);                      \
 113 } while (0)
 114 #endif
 115
 116 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
 117
 118 /* TUN device flags */
 119
 120 /* IFF_ATTACH_QUEUE is never stored in device flags,
 121  * overload it to mean fasync when stored there.
 122  */
 123 #define TUN_FASYNC      IFF_ATTACH_QUEUE
 124 /* High bits in flags field are unused. */
 125 #define TUN_VNET_LE     0x80000000
 126 #define TUN_VNET_BE     0x40000000
 127
 128 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 129                       IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
 130
 131 #define GOODCOPY_LEN 128
 132
 133 #define FLT_EXACT_COUNT 8
 134 struct tap_filter {
 135         unsigned int    count;    /* Number of addrs. Zero means disabled */
 136         u32             mask[2];  /* Mask of the hashed addrs */
 137         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 138 };
 139
 140 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 141  * to max number of VCPUs in guest. */
 142 #define MAX_TAP_QUEUES 256
 143 #define MAX_TAP_FLOWS  4096
 144
 145 #define TUN_FLOW_EXPIRE (3 * HZ)
 146
 147 struct tun_pcpu_stats {
 148         u64 rx_packets;
 149         u64 rx_bytes;
 150         u64 tx_packets;
 151         u64 tx_bytes;
 152         struct u64_stats_sync syncp;
 153         u32 rx_dropped;
 154         u32 tx_dropped;
 155         u32 rx_frame_errors;
 156 };
 157
 158 /* A tun_file connects an open character device to a tuntap netdevice. It
 159  * also contains all socket related structures (except sock_fprog and tap_filter)
 160  * to serve as one transmit queue for tuntap device. The sock_fprog and
 161  * tap_filter were kept in tun_struct since they were used for filtering for the
 162  * netdevice not for a specific queue (at least I didn't see the requirement for
 163  * this).
 164  *
 165  * RCU usage:
 166  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 167  * other can only be read while rcu_read_lock or rtnl_lock is held.
 168  */
 169 struct tun_file {
 170         struct sock sk;
 171         struct socket socket;
 172         struct socket_wq wq;
 173         struct tun_struct __rcu *tun;
 174         struct fasync_struct *fasync;
 175         /* only used for fasnyc */
 176         unsigned int flags;
 177         union {
 178                 u16 queue_index;
 179                 unsigned int ifindex;
 180         };
 181         struct napi_struct napi;
 182         bool napi_enabled;
 183         bool napi_frags_enabled;
 184         struct mutex napi_mutex;        /* Protects access to the above napi */
 185         struct list_head next;
 186         struct tun_struct *detached;
 187         struct ptr_ring tx_ring;
 188         struct xdp_rxq_info xdp_rxq;
 189 };
 190
 191 struct tun_page {
 192         struct page *page;
 193         int count;
 194 };
 195
 196 struct tun_flow_entry {
 197         struct hlist_node hash_link;
 198         struct rcu_head rcu;
 199         struct tun_struct *tun;
 200
 201         u32 rxhash;
 202         u32 rps_rxhash;
 203         int queue_index;
 204         unsigned long updated ____cacheline_aligned_in_smp;
 205 };
 206
 207 #define TUN_NUM_FLOW_ENTRIES 1024
 208 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
 209
 210 struct tun_prog {
 211         struct rcu_head rcu;
 212         struct bpf_prog *prog;
 213 };
 214
 215 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 216  * device, socket filter, sndbuf and vnet header size were restore when the
 217  * file were attached to a persist device.
 218  */
 219 struct tun_struct {
 220         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 221         unsigned int            numqueues;
 222         unsigned int            flags;
 223         kuid_t                  owner;
 224         kgid_t                  group;
 225
 226         struct net_device       *dev;
 227         netdev_features_t       set_features;
 228 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 229                           NETIF_F_TSO6)
 230
 231         int                     align;
 232         int                     vnet_hdr_sz;
 233         int                     sndbuf;
 234         struct tap_filter       txflt;
 235         struct sock_fprog       fprog;
 236         /* protected by rtnl lock */
 237         bool                    filter_attached;
 238 #ifdef TUN_DEBUG
 239         int debug;
 240 #endif
 241         spinlock_t lock;
 242         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 243         struct timer_list flow_gc_timer;
 244         unsigned long ageing_time;
 245         unsigned int numdisabled;
 246         struct list_head disabled;
 247         void *security;
 248         u32 flow_count;
 249         u32 rx_batched;
 250         struct tun_pcpu_stats __percpu *pcpu_stats;
 251         struct bpf_prog __rcu *xdp_prog;
 252         struct tun_prog __rcu *steering_prog;
 253         struct tun_prog __rcu *filter_prog;
 254         struct ethtool_link_ksettings link_ksettings;
 255 };
 256
 257 struct veth {
 258         __be16 h_vlan_proto;
 259         __be16 h_vlan_TCI;
 260 };
 261
 262 bool tun_is_xdp_frame(void *ptr)
 263 {
 264         return (unsigned long)ptr & TUN_XDP_FLAG;
 265 }
 266 EXPORT_SYMBOL(tun_is_xdp_frame);
 267
 268 void *tun_xdp_to_ptr(void *ptr)
 269 {
 270         return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
 271 }
 272 EXPORT_SYMBOL(tun_xdp_to_ptr);
 273
 274 void *tun_ptr_to_xdp(void *ptr)
 275 {
 276         return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
 277 }
 278 EXPORT_SYMBOL(tun_ptr_to_xdp);
 279
 280 static int tun_napi_receive(struct napi_struct *napi, int budget)
 281 {
 282         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 283         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 284         struct sk_buff_head process_queue;
 285         struct sk_buff *skb;
 286         int received = 0;
 287
 288         __skb_queue_head_init(&process_queue);
 289
 290         spin_lock(&queue->lock);
 291         skb_queue_splice_tail_init(queue, &process_queue);
 292         spin_unlock(&queue->lock);
 293
 294         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 295                 napi_gro_receive(napi, skb);
 296                 ++received;
 297         }
 298
 299         if (!skb_queue_empty(&process_queue)) {
 300                 spin_lock(&queue->lock);
 301                 skb_queue_splice(&process_queue, queue);
 302                 spin_unlock(&queue->lock);
 303         }
 304
 305         return received;
 306 }
 307
 308 static int tun_napi_poll(struct napi_struct *napi, int budget)
 309 {
 310         unsigned int received;
 311
 312         received = tun_napi_receive(napi, budget);
 313
 314         if (received < budget)
 315                 napi_complete_done(napi, received);
 316
 317         return received;
 318 }
 319
 320 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 321                           bool napi_en, bool napi_frags)
 322 {
 323         tfile->napi_enabled = napi_en;
 324         tfile->napi_frags_enabled = napi_en && napi_frags;
 325         if (napi_en) {
 326                 netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
 327                                NAPI_POLL_WEIGHT);
 328                 napi_enable(&tfile->napi);
 329         }
 330 }
 331
 332 static void tun_napi_disable(struct tun_file *tfile)
 333 {
 334         if (tfile->napi_enabled)
 335                 napi_disable(&tfile->napi);
 336 }
 337
 338 static void tun_napi_del(struct tun_file *tfile)
 339 {
 340         if (tfile->napi_enabled)
 341                 netif_napi_del(&tfile->napi);
 342 }
 343
 344 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
 345 {
 346         return tfile->napi_frags_enabled;
 347 }
 348
 349 #ifdef CONFIG_TUN_VNET_CROSS_LE
 350 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 351 {
 352         return tun->flags & TUN_VNET_BE ? false :
 353                 virtio_legacy_is_little_endian();
 354 }
 355
 356 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 357 {
 358         int be = !!(tun->flags & TUN_VNET_BE);
 359
 360         if (put_user(be, argp))
 361                 return -EFAULT;
 362
 363         return 0;
 364 }
 365
 366 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 367 {
 368         int be;
 369
 370         if (get_user(be, argp))
 371                 return -EFAULT;
 372
 373         if (be)
 374                 tun->flags |= TUN_VNET_BE;
 375         else
 376                 tun->flags &= ~TUN_VNET_BE;
 377
 378         return 0;
 379 }
 380 #else
 381 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 382 {
 383         return virtio_legacy_is_little_endian();
 384 }
 385
 386 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 387 {
 388         return -EINVAL;
 389 }
 390
 391 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 392 {
 393         return -EINVAL;
 394 }
 395 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 396
 397 static inline bool tun_is_little_endian(struct tun_struct *tun)
 398 {
 399         return tun->flags & TUN_VNET_LE ||
 400                 tun_legacy_is_little_endian(tun);
 401 }
 402
 403 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 404 {
 405         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 406 }
 407
 408 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 409 {
 410         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 411 }
 412
 413 static inline u32 tun_hashfn(u32 rxhash)
 414 {
 415         return rxhash & TUN_MASK_FLOW_ENTRIES;
 416 }
 417
 418 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 419 {
 420         struct tun_flow_entry *e;
 421
 422         hlist_for_each_entry_rcu(e, head, hash_link) {
 423                 if (e->rxhash == rxhash)
 424                         return e;
 425         }
 426         return NULL;
 427 }
 428
 429 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 430                                               struct hlist_head *head,
 431                                               u32 rxhash, u16 queue_index)
 432 {
 433         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 434
 435         if (e) {
 436                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 437                           rxhash, queue_index);
 438                 e->updated = jiffies;
 439                 e->rxhash = rxhash;
 440                 e->rps_rxhash = 0;
 441                 e->queue_index = queue_index;
 442                 e->tun = tun;
 443                 hlist_add_head_rcu(&e->hash_link, head);
 444                 ++tun->flow_count;
 445         }
 446         return e;
 447 }
 448
 449 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 450 {
 451         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 452                   e->rxhash, e->queue_index);
 453         hlist_del_rcu(&e->hash_link);
 454         kfree_rcu(e, rcu);
 455         --tun->flow_count;
 456 }
 457
 458 static void tun_flow_flush(struct tun_struct *tun)
 459 {
 460         int i;
 461
 462         spin_lock_bh(&tun->lock);
 463         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 464                 struct tun_flow_entry *e;
 465                 struct hlist_node *n;
 466
 467                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 468                         tun_flow_delete(tun, e);
 469         }
 470         spin_unlock_bh(&tun->lock);
 471 }
 472
 473 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 474 {
 475         int i;
 476
 477         spin_lock_bh(&tun->lock);
 478         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 479                 struct tun_flow_entry *e;
 480                 struct hlist_node *n;
 481
 482                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 483                         if (e->queue_index == queue_index)
 484                                 tun_flow_delete(tun, e);
 485                 }
 486         }
 487         spin_unlock_bh(&tun->lock);
 488 }
 489
 490 static void tun_flow_cleanup(struct timer_list *t)
 491 {
 492         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 493         unsigned long delay = tun->ageing_time;
 494         unsigned long next_timer = jiffies + delay;
 495         unsigned long count = 0;
 496         int i;
 497
 498         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 499
 500         spin_lock(&tun->lock);
 501         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 502                 struct tun_flow_entry *e;
 503                 struct hlist_node *n;
 504
 505                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 506                         unsigned long this_timer;
 507
 508                         this_timer = e->updated + delay;
 509                         if (time_before_eq(this_timer, jiffies)) {
 510                                 tun_flow_delete(tun, e);
 511                                 continue;
 512                         }
 513                         count++;
 514                         if (time_before(this_timer, next_timer))
 515                                 next_timer = this_timer;
 516                 }
 517         }
 518
 519         if (count)
 520                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 521         spin_unlock(&tun->lock);
 522 }
 523
 524 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 525                             struct tun_file *tfile)
 526 {
 527         struct hlist_head *head;
 528         struct tun_flow_entry *e;
 529         unsigned long delay = tun->ageing_time;
 530         u16 queue_index = tfile->queue_index;
 531
 532         head = &tun->flows[tun_hashfn(rxhash)];
 533
 534         rcu_read_lock();
 535
 536         e = tun_flow_find(head, rxhash);
 537         if (likely(e)) {
 538                 /* TODO: keep queueing to old queue until it's empty? */
 539                 if (e->queue_index != queue_index)
 540                         e->queue_index = queue_index;
 541                 if (e->updated != jiffies)
 542                         e->updated = jiffies;
 543                 sock_rps_record_flow_hash(e->rps_rxhash);
 544         } else {
 545                 spin_lock_bh(&tun->lock);
 546                 if (!tun_flow_find(head, rxhash) &&
 547                     tun->flow_count < MAX_TAP_FLOWS)
 548                         tun_flow_create(tun, head, rxhash, queue_index);
 549
 550                 if (!timer_pending(&tun->flow_gc_timer))
 551                         mod_timer(&tun->flow_gc_timer,
 552                                   round_jiffies_up(jiffies + delay));
 553                 spin_unlock_bh(&tun->lock);
 554         }
 555
 556         rcu_read_unlock();
 557 }
 558
 559 /**
 560  * Save the hash received in the stack receive path and update the
 561  * flow_hash table accordingly.
 562  */
 563 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 564 {
 565         if (unlikely(e->rps_rxhash != hash))
 566                 e->rps_rxhash = hash;
 567 }
 568
 569 /* We try to identify a flow through its rxhash. The reason that
 570  * we do not check rxq no. is because some cards(e.g 82599), chooses
 571  * the rxq based on the txq where the last packet of the flow comes. As
 572  * the userspace application move between processors, we may get a
 573  * different rxq no. here.
 574  */
 575 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 576 {
 577         struct tun_flow_entry *e;
 578         u32 txq = 0;
 579         u32 numqueues = 0;
 580
 581         numqueues = READ_ONCE(tun->numqueues);
 582
 583         txq = __skb_get_hash_symmetric(skb);
 584         e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 585         if (e) {
 586                 tun_flow_save_rps_rxhash(e, txq);
 587                 txq = e->queue_index;
 588         } else {
 589                 /* use multiply and shift instead of expensive divide */
 590                 txq = ((u64)txq * numqueues) >> 32;
 591         }
 592
 593         return txq;
 594 }
 595
 596 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 597 {
 598         struct tun_prog *prog;
 599         u32 numqueues;
 600         u16 ret = 0;
 601
 602         numqueues = READ_ONCE(tun->numqueues);
 603         if (!numqueues)
 604                 return 0;
 605
 606         prog = rcu_dereference(tun->steering_prog);
 607         if (prog)
 608                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 609
 610         return ret % numqueues;
 611 }
 612
 613 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 614                             struct net_device *sb_dev)
 615 {
 616         struct tun_struct *tun = netdev_priv(dev);
 617         u16 ret;
 618
 619         rcu_read_lock();
 620         if (rcu_dereference(tun->steering_prog))
 621                 ret = tun_ebpf_select_queue(tun, skb);
 622         else
 623                 ret = tun_automq_select_queue(tun, skb);
 624         rcu_read_unlock();
 625
 626         return ret;
 627 }
 628
 629 static inline bool tun_not_capable(struct tun_struct *tun)
 630 {
 631         const struct cred *cred = current_cred();
 632         struct net *net = dev_net(tun->dev);
 633
 634         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 635                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 636                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 637 }
 638
 639 static void tun_set_real_num_queues(struct tun_struct *tun)
 640 {
 641         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 642         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 643 }
 644
 645 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 646 {
 647         tfile->detached = tun;
 648         list_add_tail(&tfile->next, &tun->disabled);
 649         ++tun->numdisabled;
 650 }
 651
 652 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 653 {
 654         struct tun_struct *tun = tfile->detached;
 655
 656         tfile->detached = NULL;
 657         list_del_init(&tfile->next);
 658         --tun->numdisabled;
 659         return tun;
 660 }
 661
 662 void tun_ptr_free(void *ptr)
 663 {
 664         if (!ptr)
 665                 return;
 666         if (tun_is_xdp_frame(ptr)) {
 667                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
 668
 669                 xdp_return_frame(xdpf);
 670         } else {
 671                 __skb_array_destroy_skb(ptr);
 672         }
 673 }
 674 EXPORT_SYMBOL_GPL(tun_ptr_free);
 675
 676 static void tun_queue_purge(struct tun_file *tfile)
 677 {
 678         void *ptr;
 679
 680         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 681                 tun_ptr_free(ptr);
 682
 683         skb_queue_purge(&tfile->sk.sk_write_queue);
 684         skb_queue_purge(&tfile->sk.sk_error_queue);
 685 }
 686
 687 static void __tun_detach(struct tun_file *tfile, bool clean)
 688 {
 689         struct tun_file *ntfile;
 690         struct tun_struct *tun;
 691
 692         tun = rtnl_dereference(tfile->tun);
 693
 694         if (tun && clean) {
 695                 tun_napi_disable(tfile);
 696                 tun_napi_del(tfile);
 697         }
 698
 699         if (tun && !tfile->detached) {
 700                 u16 index = tfile->queue_index;
 701                 BUG_ON(index >= tun->numqueues);
 702
 703                 rcu_assign_pointer(tun->tfiles[index],
 704                                    tun->tfiles[tun->numqueues - 1]);
 705                 ntfile = rtnl_dereference(tun->tfiles[index]);
 706                 ntfile->queue_index = index;
 707                 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
 708                                    NULL);
 709
 710                 --tun->numqueues;
 711                 if (clean) {
 712                         RCU_INIT_POINTER(tfile->tun, NULL);
 713                         sock_put(&tfile->sk);
 714                 } else
 715                         tun_disable_queue(tun, tfile);
 716
 717                 synchronize_net();
 718                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 719                 /* Drop read queue */
 720                 tun_queue_purge(tfile);
 721                 tun_set_real_num_queues(tun);
 722         } else if (tfile->detached && clean) {
 723                 tun = tun_enable_queue(tfile);
 724                 sock_put(&tfile->sk);
 725         }
 726
 727         if (clean) {
 728                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 729                         netif_carrier_off(tun->dev);
 730
 731                         if (!(tun->flags & IFF_PERSIST) &&
 732                             tun->dev->reg_state == NETREG_REGISTERED)
 733                                 unregister_netdevice(tun->dev);
 734                 }
 735                 if (tun)
 736                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 737                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 738                 sock_put(&tfile->sk);
 739         }
 740 }
 741
 742 static void tun_detach(struct tun_file *tfile, bool clean)
 743 {
 744         struct tun_struct *tun;
 745         struct net_device *dev;
 746
 747         rtnl_lock();
 748         tun = rtnl_dereference(tfile->tun);
 749         dev = tun ? tun->dev : NULL;
 750         __tun_detach(tfile, clean);
 751         if (dev)
 752                 netdev_state_change(dev);
 753         rtnl_unlock();
 754 }
 755
 756 static void tun_detach_all(struct net_device *dev)
 757 {
 758         struct tun_struct *tun = netdev_priv(dev);
 759         struct tun_file *tfile, *tmp;
 760         int i, n = tun->numqueues;
 761
 762         for (i = 0; i < n; i++) {
 763                 tfile = rtnl_dereference(tun->tfiles[i]);
 764                 BUG_ON(!tfile);
 765                 tun_napi_disable(tfile);
 766                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 767                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 768                 RCU_INIT_POINTER(tfile->tun, NULL);
 769                 --tun->numqueues;
 770         }
 771         list_for_each_entry(tfile, &tun->disabled, next) {
 772                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 773                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 774                 RCU_INIT_POINTER(tfile->tun, NULL);
 775         }
 776         BUG_ON(tun->numqueues != 0);
 777
 778         synchronize_net();
 779         for (i = 0; i < n; i++) {
 780                 tfile = rtnl_dereference(tun->tfiles[i]);
 781                 tun_napi_del(tfile);
 782                 /* Drop read queue */
 783                 tun_queue_purge(tfile);
 784                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 785                 sock_put(&tfile->sk);
 786         }
 787         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 788                 tun_enable_queue(tfile);
 789                 tun_queue_purge(tfile);
 790                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 791                 sock_put(&tfile->sk);
 792         }
 793         BUG_ON(tun->numdisabled != 0);
 794
 795         if (tun->flags & IFF_PERSIST)
 796                 module_put(THIS_MODULE);
 797 }
 798
 799 static int tun_attach(struct tun_struct *tun, struct file *file,
 800                       bool skip_filter, bool napi, bool napi_frags)
 801 {
 802         struct tun_file *tfile = file->private_data;
 803         struct net_device *dev = tun->dev;
 804         int err;
 805
 806         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 807         if (err < 0)
 808                 goto out;
 809
 810         err = -EINVAL;
 811         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 812                 goto out;
 813
 814         err = -EBUSY;
 815         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 816                 goto out;
 817
 818         err = -E2BIG;
 819         if (!tfile->detached &&
 820             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 821                 goto out;
 822
 823         err = 0;
 824
 825         /* Re-attach the filter to persist device */
 826         if (!skip_filter && (tun->filter_attached == true)) {
 827                 lock_sock(tfile->socket.sk);
 828                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 829                 release_sock(tfile->socket.sk);
 830                 if (!err)
 831                         goto out;
 832         }
 833
 834         if (!tfile->detached &&
 835             ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
 836                             GFP_KERNEL, tun_ptr_free)) {
 837                 err = -ENOMEM;
 838                 goto out;
 839         }
 840
 841         tfile->queue_index = tun->numqueues;
 842         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 843
 844         if (tfile->detached) {
 845                 /* Re-attach detached tfile, updating XDP queue_index */
 846                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 847
 848                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 849                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 850         } else {
 851                 /* Setup XDP RX-queue info, for new tfile getting attached */
 852                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 853                                        tun->dev, tfile->queue_index);
 854                 if (err < 0)
 855                         goto out;
 856                 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 857                                                  MEM_TYPE_PAGE_SHARED, NULL);
 858                 if (err < 0) {
 859                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 860                         goto out;
 861                 }
 862                 err = 0;
 863         }
 864
 865         if (tfile->detached) {
 866                 tun_enable_queue(tfile);
 867         } else {
 868                 sock_hold(&tfile->sk);
 869                 tun_napi_init(tun, tfile, napi, napi_frags);
 870         }
 871
 872         if (rtnl_dereference(tun->xdp_prog))
 873                 sock_set_flag(&tfile->sk, SOCK_XDP);
 874
 875         /* device is allowed to go away first, so no need to hold extra
 876          * refcnt.
 877          */
 878
 879         /* Publish tfile->tun and tun->tfiles only after we've fully
 880          * initialized tfile; otherwise we risk using half-initialized
 881          * object.
 882          */
 883         rcu_assign_pointer(tfile->tun, tun);
 884         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 885         tun->numqueues++;
 886         tun_set_real_num_queues(tun);
 887 out:
 888         return err;
 889 }
 890
 891 static struct tun_struct *tun_get(struct tun_file *tfile)
 892 {
 893         struct tun_struct *tun;
 894
 895         rcu_read_lock();
 896         tun = rcu_dereference(tfile->tun);
 897         if (tun)
 898                 dev_hold(tun->dev);
 899         rcu_read_unlock();
 900
 901         return tun;
 902 }
 903
 904 static void tun_put(struct tun_struct *tun)
 905 {
 906         dev_put(tun->dev);
 907 }
 908
 909 /* TAP filtering */
 910 static void addr_hash_set(u32 *mask, const u8 *addr)
 911 {
 912         int n = ether_crc(ETH_ALEN, addr) >> 26;
 913         mask[n >> 5] |= (1 << (n & 31));
 914 }
 915
 916 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 917 {
 918         int n = ether_crc(ETH_ALEN, addr) >> 26;
 919         return mask[n >> 5] & (1 << (n & 31));
 920 }
 921
 922 static int update_filter(struct tap_filter *filter, void __user *arg)
 923 {
 924         struct { u8 u[ETH_ALEN]; } *addr;
 925         struct tun_filter uf;
 926         int err, alen, n, nexact;
 927
 928         if (copy_from_user(&uf, arg, sizeof(uf)))
 929                 return -EFAULT;
 930
 931         if (!uf.count) {
 932                 /* Disabled */
 933                 filter->count = 0;
 934                 return 0;
 935         }
 936
 937         alen = ETH_ALEN * uf.count;
 938         addr = memdup_user(arg + sizeof(uf), alen);
 939         if (IS_ERR(addr))
 940                 return PTR_ERR(addr);
 941
 942         /* The filter is updated without holding any locks. Which is
 943          * perfectly safe. We disable it first and in the worst
 944          * case we'll accept a few undesired packets. */
 945         filter->count = 0;
 946         wmb();
 947
 948         /* Use first set of addresses as an exact filter */
 949         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 950                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 951
 952         nexact = n;
 953
 954         /* Remaining multicast addresses are hashed,
 955          * unicast will leave the filter disabled. */
 956         memset(filter->mask, 0, sizeof(filter->mask));
 957         for (; n < uf.count; n++) {
 958                 if (!is_multicast_ether_addr(addr[n].u)) {
 959                         err = 0; /* no filter */
 960                         goto free_addr;
 961                 }
 962                 addr_hash_set(filter->mask, addr[n].u);
 963         }
 964
 965         /* For ALLMULTI just set the mask to all ones.
 966          * This overrides the mask populated above. */
 967         if ((uf.flags & TUN_FLT_ALLMULTI))
 968                 memset(filter->mask, ~0, sizeof(filter->mask));
 969
 970         /* Now enable the filter */
 971         wmb();
 972         filter->count = nexact;
 973
 974         /* Return the number of exact filters */
 975         err = nexact;
 976 free_addr:
 977         kfree(addr);
 978         return err;
 979 }
 980
 981 /* Returns: 0 - drop, !=0 - accept */
 982 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 983 {
 984         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 985          * at this point. */
 986         struct ethhdr *eh = (struct ethhdr *) skb->data;
 987         int i;
 988
 989         /* Exact match */
 990         for (i = 0; i < filter->count; i++)
 991                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 992                         return 1;
 993
 994         /* Inexact match (multicast only) */
 995         if (is_multicast_ether_addr(eh->h_dest))
 996                 return addr_hash_test(filter->mask, eh->h_dest);
 997
 998         return 0;
 999 }
1000
1001 /*
1002  * Checks whether the packet is accepted or not.
1003  * Returns: 0 - drop, !=0 - accept
1004  */
1005 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
1006 {
1007         if (!filter->count)
1008                 return 1;
1009
1010         return run_filter(filter, skb);
1011 }
1012
1013 /* Network device part of the driver */
1014
1015 static const struct ethtool_ops tun_ethtool_ops;
1016
1017 /* Net device detach from fd. */
1018 static void tun_net_uninit(struct net_device *dev)
1019 {
1020         tun_detach_all(dev);
1021 }
1022
1023 /* Net device open. */
1024 static int tun_net_open(struct net_device *dev)
1025 {
1026         struct tun_struct *tun = netdev_priv(dev);
1027         int i;
1028
1029         netif_tx_start_all_queues(dev);
1030
1031         for (i = 0; i < tun->numqueues; i++) {
1032                 struct tun_file *tfile;
1033
1034                 tfile = rtnl_dereference(tun->tfiles[i]);
1035                 tfile->socket.sk->sk_write_space(tfile->socket.sk);
1036         }
1037
1038         return 0;
1039 }
1040
1041 /* Net device close. */
1042 static int tun_net_close(struct net_device *dev)
1043 {
1044         netif_tx_stop_all_queues(dev);
1045         return 0;
1046 }
1047
1048 /* Net device start xmit */
1049 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1050 {
1051 #ifdef CONFIG_RPS
1052         if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
1053                 /* Select queue was not called for the skbuff, so we extract the
1054                  * RPS hash and save it into the flow_table here.
1055                  */
1056                 struct tun_flow_entry *e;
1057                 __u32 rxhash;
1058
1059                 rxhash = __skb_get_hash_symmetric(skb);
1060                 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1061                 if (e)
1062                         tun_flow_save_rps_rxhash(e, rxhash);
1063         }
1064 #endif
1065 }
1066
1067 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1068                                     struct sk_buff *skb,
1069                                     int len)
1070 {
1071         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1072
1073         if (prog)
1074                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1075
1076         return len;
1077 }
1078
1079 /* Net device start xmit */
1080 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1081 {
1082         struct tun_struct *tun = netdev_priv(dev);
1083         int txq = skb->queue_mapping;
1084         struct tun_file *tfile;
1085         int len = skb->len;
1086
1087         rcu_read_lock();
1088         tfile = rcu_dereference(tun->tfiles[txq]);
1089
1090         /* Drop packet if interface is not attached */
1091         if (!tfile)
1092                 goto drop;
1093
1094         if (!rcu_dereference(tun->steering_prog))
1095                 tun_automq_xmit(tun, skb);
1096
1097         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1098
1099         BUG_ON(!tfile);
1100
1101         /* Drop if the filter does not like it.
1102          * This is a noop if the filter is disabled.
1103          * Filter can be enabled only for the TAP devices. */
1104         if (!check_filter(&tun->txflt, skb))
1105                 goto drop;
1106
1107         if (tfile->socket.sk->sk_filter &&
1108             sk_filter(tfile->socket.sk, skb))
1109                 goto drop;
1110
1111         len = run_ebpf_filter(tun, skb, len);
1112         if (len == 0 || pskb_trim(skb, len))
1113                 goto drop;
1114
1115         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1116                 goto drop;
1117
1118         skb_tx_timestamp(skb);
1119
1120         /* Orphan the skb - required as we might hang on to it
1121          * for indefinite time.
1122          */
1123         skb_orphan(skb);
1124
1125         nf_reset(skb);
1126
1127         if (ptr_ring_produce(&tfile->tx_ring, skb))
1128                 goto drop;
1129
1130         /* Notify and wake up reader process */
1131         if (tfile->flags & TUN_FASYNC)
1132                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1133         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1134
1135         rcu_read_unlock();
1136         return NETDEV_TX_OK;
1137
1138 drop:
1139         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1140         skb_tx_error(skb);
1141         kfree_skb(skb);
1142         rcu_read_unlock();
1143         return NET_XMIT_DROP;
1144 }
1145
1146 static void tun_net_mclist(struct net_device *dev)
1147 {
1148         /*
1149          * This callback is supposed to deal with mc filter in
1150          * _rx_ path and has nothing to do with the _tx_ path.
1151          * In rx path we always accept everything userspace gives us.
1152          */
1153 }
1154
1155 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1156         netdev_features_t features)
1157 {
1158         struct tun_struct *tun = netdev_priv(dev);
1159
1160         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1161 }
1162
1163 static void tun_set_headroom(struct net_device *dev, int new_hr)
1164 {
1165         struct tun_struct *tun = netdev_priv(dev);
1166
1167         if (new_hr < NET_SKB_PAD)
1168                 new_hr = NET_SKB_PAD;
1169
1170         tun->align = new_hr;
1171 }
1172
1173 static void
1174 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1175 {
1176         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1177         struct tun_struct *tun = netdev_priv(dev);
1178         struct tun_pcpu_stats *p;
1179         int i;
1180
1181         for_each_possible_cpu(i) {
1182                 u64 rxpackets, rxbytes, txpackets, txbytes;
1183                 unsigned int start;
1184
1185                 p = per_cpu_ptr(tun->pcpu_stats, i);
1186                 do {
1187                         start = u64_stats_fetch_begin(&p->syncp);
1188                         rxpackets       = p->rx_packets;
1189                         rxbytes         = p->rx_bytes;
1190                         txpackets       = p->tx_packets;
1191                         txbytes         = p->tx_bytes;
1192                 } while (u64_stats_fetch_retry(&p->syncp, start));
1193
1194                 stats->rx_packets       += rxpackets;
1195                 stats->rx_bytes         += rxbytes;
1196                 stats->tx_packets       += txpackets;
1197                 stats->tx_bytes         += txbytes;
1198
1199                 /* u32 counters */
1200                 rx_dropped      += p->rx_dropped;
1201                 rx_frame_errors += p->rx_frame_errors;
1202                 tx_dropped      += p->tx_dropped;
1203         }
1204         stats->rx_dropped  = rx_dropped;
1205         stats->rx_frame_errors = rx_frame_errors;
1206         stats->tx_dropped = tx_dropped;
1207 }
1208
1209 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1210                        struct netlink_ext_ack *extack)
1211 {
1212         struct tun_struct *tun = netdev_priv(dev);
1213         struct tun_file *tfile;
1214         struct bpf_prog *old_prog;
1215         int i;
1216
1217         old_prog = rtnl_dereference(tun->xdp_prog);
1218         rcu_assign_pointer(tun->xdp_prog, prog);
1219         if (old_prog)
1220                 bpf_prog_put(old_prog);
1221
1222         for (i = 0; i < tun->numqueues; i++) {
1223                 tfile = rtnl_dereference(tun->tfiles[i]);
1224                 if (prog)
1225                         sock_set_flag(&tfile->sk, SOCK_XDP);
1226                 else
1227                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1228         }
1229         list_for_each_entry(tfile, &tun->disabled, next) {
1230                 if (prog)
1231                         sock_set_flag(&tfile->sk, SOCK_XDP);
1232                 else
1233                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1234         }
1235
1236         return 0;
1237 }
1238
1239 static u32 tun_xdp_query(struct net_device *dev)
1240 {
1241         struct tun_struct *tun = netdev_priv(dev);
1242         const struct bpf_prog *xdp_prog;
1243
1244         xdp_prog = rtnl_dereference(tun->xdp_prog);
1245         if (xdp_prog)
1246                 return xdp_prog->aux->id;
1247
1248         return 0;
1249 }
1250
1251 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1252 {
1253         switch (xdp->command) {
1254         case XDP_SETUP_PROG:
1255                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1256         case XDP_QUERY_PROG:
1257                 xdp->prog_id = tun_xdp_query(dev);
1258                 return 0;
1259         default:
1260                 return -EINVAL;
1261         }
1262 }
1263
1264 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1265 {
1266         if (new_carrier) {
1267                 struct tun_struct *tun = netdev_priv(dev);
1268
1269                 if (!tun->numqueues)
1270                         return -EPERM;
1271
1272                 netif_carrier_on(dev);
1273         } else {
1274                 netif_carrier_off(dev);
1275         }
1276         return 0;
1277 }
1278
1279 static const struct net_device_ops tun_netdev_ops = {
1280         .ndo_uninit             = tun_net_uninit,
1281         .ndo_open               = tun_net_open,
1282         .ndo_stop               = tun_net_close,
1283         .ndo_start_xmit         = tun_net_xmit,
1284         .ndo_fix_features       = tun_net_fix_features,
1285         .ndo_select_queue       = tun_select_queue,
1286         .ndo_set_rx_headroom    = tun_set_headroom,
1287         .ndo_get_stats64        = tun_net_get_stats64,
1288         .ndo_change_carrier     = tun_net_change_carrier,
1289 };
1290
1291 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1292 {
1293         /* Notify and wake up reader process */
1294         if (tfile->flags & TUN_FASYNC)
1295                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1296         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1297 }
1298
1299 static int tun_xdp_xmit(struct net_device *dev, int n,
1300                         struct xdp_frame **frames, u32 flags)
1301 {
1302         struct tun_struct *tun = netdev_priv(dev);
1303         struct tun_file *tfile;
1304         u32 numqueues;
1305         int drops = 0;
1306         int cnt = n;
1307         int i;
1308
1309         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1310                 return -EINVAL;
1311
1312         rcu_read_lock();
1313
1314 resample:
1315         numqueues = READ_ONCE(tun->numqueues);
1316         if (!numqueues) {
1317                 rcu_read_unlock();
1318                 return -ENXIO; /* Caller will free/return all frames */
1319         }
1320
1321         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1322                                             numqueues]);
1323         if (unlikely(!tfile))
1324                 goto resample;
1325
1326         spin_lock(&tfile->tx_ring.producer_lock);
1327         for (i = 0; i < n; i++) {
1328                 struct xdp_frame *xdp = frames[i];
1329                 /* Encode the XDP flag into lowest bit for consumer to differ
1330                  * XDP buffer from sk_buff.
1331                  */
1332                 void *frame = tun_xdp_to_ptr(xdp);
1333
1334                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1335                         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1336                         xdp_return_frame_rx_napi(xdp);
1337                         drops++;
1338                 }
1339         }
1340         spin_unlock(&tfile->tx_ring.producer_lock);
1341
1342         if (flags & XDP_XMIT_FLUSH)
1343                 __tun_xdp_flush_tfile(tfile);
1344
1345         rcu_read_unlock();
1346         return cnt - drops;
1347 }
1348
1349 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1350 {
1351         struct xdp_frame *frame = convert_to_xdp_frame(xdp);
1352
1353         if (unlikely(!frame))
1354                 return -EOVERFLOW;
1355
1356         return tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1357 }
1358
1359 static const struct net_device_ops tap_netdev_ops = {
1360         .ndo_uninit             = tun_net_uninit,
1361         .ndo_open               = tun_net_open,
1362         .ndo_stop               = tun_net_close,
1363         .ndo_start_xmit         = tun_net_xmit,
1364         .ndo_fix_features       = tun_net_fix_features,
1365         .ndo_set_rx_mode        = tun_net_mclist,
1366         .ndo_set_mac_address    = eth_mac_addr,
1367         .ndo_validate_addr      = eth_validate_addr,
1368         .ndo_select_queue       = tun_select_queue,
1369         .ndo_features_check     = passthru_features_check,
1370         .ndo_set_rx_headroom    = tun_set_headroom,
1371         .ndo_get_stats64        = tun_net_get_stats64,
1372         .ndo_bpf                = tun_xdp,
1373         .ndo_xdp_xmit           = tun_xdp_xmit,
1374         .ndo_change_carrier     = tun_net_change_carrier,
1375 };
1376
1377 static void tun_flow_init(struct tun_struct *tun)
1378 {
1379         int i;
1380
1381         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1382                 INIT_HLIST_HEAD(&tun->flows[i]);
1383
1384         tun->ageing_time = TUN_FLOW_EXPIRE;
1385         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1386         mod_timer(&tun->flow_gc_timer,
1387                   round_jiffies_up(jiffies + tun->ageing_time));
1388 }
1389
1390 static void tun_flow_uninit(struct tun_struct *tun)
1391 {
1392         del_timer_sync(&tun->flow_gc_timer);
1393         tun_flow_flush(tun);
1394 }
1395
1396 #define MIN_MTU 68
1397 #define MAX_MTU 65535
1398
1399 /* Initialize net device. */
1400 static void tun_net_init(struct net_device *dev)
1401 {
1402         struct tun_struct *tun = netdev_priv(dev);
1403
1404         switch (tun->flags & TUN_TYPE_MASK) {
1405         case IFF_TUN:
1406                 dev->netdev_ops = &tun_netdev_ops;
1407
1408                 /* Point-to-Point TUN Device */
1409                 dev->hard_header_len = 0;
1410                 dev->addr_len = 0;
1411                 dev->mtu = 1500;
1412
1413                 /* Zero header length */
1414                 dev->type = ARPHRD_NONE;
1415                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1416                 break;
1417
1418         case IFF_TAP:
1419                 dev->netdev_ops = &tap_netdev_ops;
1420                 /* Ethernet TAP Device */
1421                 ether_setup(dev);
1422                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1423                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1424
1425                 eth_hw_addr_random(dev);
1426
1427                 break;
1428         }
1429
1430         dev->min_mtu = MIN_MTU;
1431         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1432 }
1433
1434 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1435 {
1436         struct sock *sk = tfile->socket.sk;
1437
1438         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1439 }
1440
1441 /* Character device part */
1442
1443 /* Poll */
1444 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1445 {
1446         struct tun_file *tfile = file->private_data;
1447         struct tun_struct *tun = tun_get(tfile);
1448         struct sock *sk;
1449         __poll_t mask = 0;
1450
1451         if (!tun)
1452                 return EPOLLERR;
1453
1454         sk = tfile->socket.sk;
1455
1456         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1457
1458         poll_wait(file, sk_sleep(sk), wait);
1459
1460         if (!ptr_ring_empty(&tfile->tx_ring))
1461                 mask |= EPOLLIN | EPOLLRDNORM;
1462
1463         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1464          * guarantee EPOLLOUT to be raised by either here or
1465          * tun_sock_write_space(). Then process could get notification
1466          * after it writes to a down device and meets -EIO.
1467          */
1468         if (tun_sock_writeable(tun, tfile) ||
1469             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1470              tun_sock_writeable(tun, tfile)))
1471                 mask |= EPOLLOUT | EPOLLWRNORM;
1472
1473         if (tun->dev->reg_state != NETREG_REGISTERED)
1474                 mask = EPOLLERR;
1475
1476         tun_put(tun);
1477         return mask;
1478 }
1479
1480 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1481                                             size_t len,
1482                                             const struct iov_iter *it)
1483 {
1484         struct sk_buff *skb;
1485         size_t linear;
1486         int err;
1487         int i;
1488
1489         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1490                 return ERR_PTR(-ENOMEM);
1491
1492         local_bh_disable();
1493         skb = napi_get_frags(&tfile->napi);
1494         local_bh_enable();
1495         if (!skb)
1496                 return ERR_PTR(-ENOMEM);
1497
1498         linear = iov_iter_single_seg_count(it);
1499         err = __skb_grow(skb, linear);
1500         if (err)
1501                 goto free;
1502
1503         skb->len = len;
1504         skb->data_len = len - linear;
1505         skb->truesize += skb->data_len;
1506
1507         for (i = 1; i < it->nr_segs; i++) {
1508                 size_t fragsz = it->iov[i].iov_len;
1509                 struct page *page;
1510                 void *frag;
1511
1512                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1513                         err = -EINVAL;
1514                         goto free;
1515                 }
1516                 frag = netdev_alloc_frag(fragsz);
1517                 if (!frag) {
1518                         err = -ENOMEM;
1519                         goto free;
1520                 }
1521                 page = virt_to_head_page(frag);
1522                 skb_fill_page_desc(skb, i - 1, page,
1523                                    frag - page_address(page), fragsz);
1524         }
1525
1526         return skb;
1527 free:
1528         /* frees skb and all frags allocated with napi_alloc_frag() */
1529         napi_free_frags(&tfile->napi);
1530         return ERR_PTR(err);
1531 }
1532
1533 /* prepad is the amount to reserve at front.  len is length after that.
1534  * linear is a hint as to how much to copy (usually headers). */
1535 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1536                                      size_t prepad, size_t len,
1537                                      size_t linear, int noblock)
1538 {
1539         struct sock *sk = tfile->socket.sk;
1540         struct sk_buff *skb;
1541         int err;
1542
1543         /* Under a page?  Don't bother with paged skb. */
1544         if (prepad + len < PAGE_SIZE || !linear)
1545                 linear = len;
1546
1547         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1548                                    &err, 0);
1549         if (!skb)
1550                 return ERR_PTR(err);
1551
1552         skb_reserve(skb, prepad);
1553         skb_put(skb, linear);
1554         skb->data_len = len - linear;
1555         skb->len += len - linear;
1556
1557         return skb;
1558 }
1559
1560 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1561                            struct sk_buff *skb, int more)
1562 {
1563         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1564         struct sk_buff_head process_queue;
1565         u32 rx_batched = tun->rx_batched;
1566         bool rcv = false;
1567
1568         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1569                 local_bh_disable();
1570                 skb_record_rx_queue(skb, tfile->queue_index);
1571                 netif_receive_skb(skb);
1572                 local_bh_enable();
1573                 return;
1574         }
1575
1576         spin_lock(&queue->lock);
1577         if (!more || skb_queue_len(queue) == rx_batched) {
1578                 __skb_queue_head_init(&process_queue);
1579                 skb_queue_splice_tail_init(queue, &process_queue);
1580                 rcv = true;
1581         } else {
1582                 __skb_queue_tail(queue, skb);
1583         }
1584         spin_unlock(&queue->lock);
1585
1586         if (rcv) {
1587                 struct sk_buff *nskb;
1588
1589                 local_bh_disable();
1590                 while ((nskb = __skb_dequeue(&process_queue))) {
1591                         skb_record_rx_queue(nskb, tfile->queue_index);
1592                         netif_receive_skb(nskb);
1593                 }
1594                 skb_record_rx_queue(skb, tfile->queue_index);
1595                 netif_receive_skb(skb);
1596                 local_bh_enable();
1597         }
1598 }
1599
1600 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1601                               int len, int noblock, bool zerocopy)
1602 {
1603         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1604                 return false;
1605
1606         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1607                 return false;
1608
1609         if (!noblock)
1610                 return false;
1611
1612         if (zerocopy)
1613                 return false;
1614
1615         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1616             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1617                 return false;
1618
1619         return true;
1620 }
1621
1622 static struct sk_buff *__tun_build_skb(struct page_frag *alloc_frag, char *buf,
1623                                        int buflen, int len, int pad)
1624 {
1625         struct sk_buff *skb = build_skb(buf, buflen);
1626
1627         if (!skb)
1628                 return ERR_PTR(-ENOMEM);
1629
1630         skb_reserve(skb, pad);
1631         skb_put(skb, len);
1632
1633         get_page(alloc_frag->page);
1634         alloc_frag->offset += buflen;
1635
1636         return skb;
1637 }
1638
1639 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1640                        struct xdp_buff *xdp, u32 act)
1641 {
1642         int err;
1643
1644         switch (act) {
1645         case XDP_REDIRECT:
1646                 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1647                 if (err)
1648                         return err;
1649                 break;
1650         case XDP_TX:
1651                 err = tun_xdp_tx(tun->dev, xdp);
1652                 if (err < 0)
1653                         return err;
1654                 break;
1655         case XDP_PASS:
1656                 break;
1657         default:
1658                 bpf_warn_invalid_xdp_action(act);
1659                 /* fall through */
1660         case XDP_ABORTED:
1661                 trace_xdp_exception(tun->dev, xdp_prog, act);
1662                 /* fall through */
1663         case XDP_DROP:
1664                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1665                 break;
1666         }
1667
1668         return act;
1669 }
1670
1671 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1672                                      struct tun_file *tfile,
1673                                      struct iov_iter *from,
1674                                      struct virtio_net_hdr *hdr,
1675                                      int len, int *skb_xdp)
1676 {
1677         struct page_frag *alloc_frag = &current->task_frag;
1678         struct bpf_prog *xdp_prog;
1679         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1680         char *buf;
1681         size_t copied;
1682         int pad = TUN_RX_PAD;
1683         int err = 0;
1684
1685         rcu_read_lock();
1686         xdp_prog = rcu_dereference(tun->xdp_prog);
1687         if (xdp_prog)
1688                 pad += XDP_PACKET_HEADROOM;
1689         buflen += SKB_DATA_ALIGN(len + pad);
1690         rcu_read_unlock();
1691
1692         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1693         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1694                 return ERR_PTR(-ENOMEM);
1695
1696         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1697         copied = copy_page_from_iter(alloc_frag->page,
1698                                      alloc_frag->offset + pad,
1699                                      len, from);
1700         if (copied != len)
1701                 return ERR_PTR(-EFAULT);
1702
1703         /* There's a small window that XDP may be set after the check
1704          * of xdp_prog above, this should be rare and for simplicity
1705          * we do XDP on skb in case the headroom is not enough.
1706          */
1707         if (hdr->gso_type || !xdp_prog) {
1708                 *skb_xdp = 1;
1709                 return __tun_build_skb(alloc_frag, buf, buflen, len, pad);
1710         }
1711
1712         *skb_xdp = 0;
1713
1714         local_bh_disable();
1715         rcu_read_lock();
1716         xdp_prog = rcu_dereference(tun->xdp_prog);
1717         if (xdp_prog) {
1718                 struct xdp_buff xdp;
1719                 u32 act;
1720
1721                 xdp.data_hard_start = buf;
1722                 xdp.data = buf + pad;
1723                 xdp_set_data_meta_invalid(&xdp);
1724                 xdp.data_end = xdp.data + len;
1725                 xdp.rxq = &tfile->xdp_rxq;
1726
1727                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1728                 if (act == XDP_REDIRECT || act == XDP_TX) {
1729                         get_page(alloc_frag->page);
1730                         alloc_frag->offset += buflen;
1731                 }
1732                 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1733                 if (err < 0)
1734                         goto err_xdp;
1735                 if (err == XDP_REDIRECT)
1736                         xdp_do_flush_map();
1737                 if (err != XDP_PASS)
1738                         goto out;
1739
1740                 pad = xdp.data - xdp.data_hard_start;
1741                 len = xdp.data_end - xdp.data;
1742         }
1743         rcu_read_unlock();
1744         local_bh_enable();
1745
1746         return __tun_build_skb(alloc_frag, buf, buflen, len, pad);
1747
1748 err_xdp:
1749         put_page(alloc_frag->page);
1750 out:
1751         rcu_read_unlock();
1752         local_bh_enable();
1753         return NULL;
1754 }
1755
1756 /* Get packet from user space buffer */
1757 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1758                             void *msg_control, struct iov_iter *from,
1759                             int noblock, bool more)
1760 {
1761         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1762         struct sk_buff *skb;
1763         size_t total_len = iov_iter_count(from);
1764         size_t len = total_len, align = tun->align, linear;
1765         struct virtio_net_hdr gso = { 0 };
1766         struct tun_pcpu_stats *stats;
1767         int good_linear;
1768         int copylen;
1769         bool zerocopy = false;
1770         int err;
1771         u32 rxhash = 0;
1772         int skb_xdp = 1;
1773         bool frags = tun_napi_frags_enabled(tfile);
1774
1775         if (!(tun->flags & IFF_NO_PI)) {
1776                 if (len < sizeof(pi))
1777                         return -EINVAL;
1778                 len -= sizeof(pi);
1779
1780                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1781                         return -EFAULT;
1782         }
1783
1784         if (tun->flags & IFF_VNET_HDR) {
1785                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1786
1787                 if (len < vnet_hdr_sz)
1788                         return -EINVAL;
1789                 len -= vnet_hdr_sz;
1790
1791                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1792                         return -EFAULT;
1793
1794                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1795                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1796                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1797
1798                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1799                         return -EINVAL;
1800                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1801         }
1802
1803         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1804                 align += NET_IP_ALIGN;
1805                 if (unlikely(len < ETH_HLEN ||
1806                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1807                         return -EINVAL;
1808         }
1809
1810         good_linear = SKB_MAX_HEAD(align);
1811
1812         if (msg_control) {
1813                 struct iov_iter i = *from;
1814
1815                 /* There are 256 bytes to be copied in skb, so there is
1816                  * enough room for skb expand head in case it is used.
1817                  * The rest of the buffer is mapped from userspace.
1818                  */
1819                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1820                 if (copylen > good_linear)
1821                         copylen = good_linear;
1822                 linear = copylen;
1823                 iov_iter_advance(&i, copylen);
1824                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1825                         zerocopy = true;
1826         }
1827
1828         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1829                 /* For the packet that is not easy to be processed
1830                  * (e.g gso or jumbo packet), we will do it at after
1831                  * skb was created with generic XDP routine.
1832                  */
1833                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1834                 if (IS_ERR(skb)) {
1835                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1836                         return PTR_ERR(skb);
1837                 }
1838                 if (!skb)
1839                         return total_len;
1840         } else {
1841                 if (!zerocopy) {
1842                         copylen = len;
1843                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1844                                 linear = good_linear;
1845                         else
1846                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1847                 }
1848
1849                 if (frags) {
1850                         mutex_lock(&tfile->napi_mutex);
1851                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1852                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1853                          * If zerocopy is enabled, then this layout will be
1854                          * overwritten by zerocopy_sg_from_iter().
1855                          */
1856                         zerocopy = false;
1857                 } else {
1858                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1859                                             noblock);
1860                 }
1861
1862                 if (IS_ERR(skb)) {
1863                         if (PTR_ERR(skb) != -EAGAIN)
1864                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1865                         if (frags)
1866                                 mutex_unlock(&tfile->napi_mutex);
1867                         return PTR_ERR(skb);
1868                 }
1869
1870                 if (zerocopy)
1871                         err = zerocopy_sg_from_iter(skb, from);
1872                 else
1873                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1874
1875                 if (err) {
1876                         err = -EFAULT;
1877 drop:
1878                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1879                         kfree_skb(skb);
1880                         if (frags) {
1881                                 tfile->napi.skb = NULL;
1882                                 mutex_unlock(&tfile->napi_mutex);
1883                         }
1884
1885                         return err;
1886                 }
1887         }
1888
1889         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1890                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1891                 kfree_skb(skb);
1892                 if (frags) {
1893                         tfile->napi.skb = NULL;
1894                         mutex_unlock(&tfile->napi_mutex);
1895                 }
1896
1897                 return -EINVAL;
1898         }
1899
1900         switch (tun->flags & TUN_TYPE_MASK) {
1901         case IFF_TUN:
1902                 if (tun->flags & IFF_NO_PI) {
1903                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1904
1905                         switch (ip_version) {
1906                         case 4:
1907                                 pi.proto = htons(ETH_P_IP);
1908                                 break;
1909                         case 6:
1910                                 pi.proto = htons(ETH_P_IPV6);
1911                                 break;
1912                         default:
1913                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1914                                 kfree_skb(skb);
1915                                 return -EINVAL;
1916                         }
1917                 }
1918
1919                 skb_reset_mac_header(skb);
1920                 skb->protocol = pi.proto;
1921                 skb->dev = tun->dev;
1922                 break;
1923         case IFF_TAP:
1924                 if (!frags)
1925                         skb->protocol = eth_type_trans(skb, tun->dev);
1926                 break;
1927         }
1928
1929         /* copy skb_ubuf_info for callback when skb has no error */
1930         if (zerocopy) {
1931                 skb_shinfo(skb)->destructor_arg = msg_control;
1932                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1933                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1934         } else if (msg_control) {
1935                 struct ubuf_info *uarg = msg_control;
1936                 uarg->callback(uarg, false);
1937         }
1938
1939         skb_reset_network_header(skb);
1940         skb_probe_transport_header(skb);
1941
1942         if (skb_xdp) {
1943                 struct bpf_prog *xdp_prog;
1944                 int ret;
1945
1946                 local_bh_disable();
1947                 rcu_read_lock();
1948                 xdp_prog = rcu_dereference(tun->xdp_prog);
1949                 if (xdp_prog) {
1950                         ret = do_xdp_generic(xdp_prog, skb);
1951                         if (ret != XDP_PASS) {
1952                                 rcu_read_unlock();
1953                                 local_bh_enable();
1954                                 return total_len;
1955                         }
1956                 }
1957                 rcu_read_unlock();
1958                 local_bh_enable();
1959         }
1960
1961         /* Compute the costly rx hash only if needed for flow updates.
1962          * We may get a very small possibility of OOO during switching, not
1963          * worth to optimize.
1964          */
1965         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1966             !tfile->detached)
1967                 rxhash = __skb_get_hash_symmetric(skb);
1968
1969         rcu_read_lock();
1970         if (unlikely(!(tun->dev->flags & IFF_UP))) {
1971                 err = -EIO;
1972                 rcu_read_unlock();
1973                 goto drop;
1974         }
1975
1976         if (frags) {
1977                 /* Exercise flow dissector code path. */
1978                 u32 headlen = eth_get_headlen(tun->dev, skb->data,
1979                                               skb_headlen(skb));
1980
1981                 if (unlikely(headlen > skb_headlen(skb))) {
1982                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1983                         napi_free_frags(&tfile->napi);
1984                         rcu_read_unlock();
1985                         mutex_unlock(&tfile->napi_mutex);
1986                         WARN_ON(1);
1987                         return -ENOMEM;
1988                 }
1989
1990                 local_bh_disable();
1991                 napi_gro_frags(&tfile->napi);
1992                 local_bh_enable();
1993                 mutex_unlock(&tfile->napi_mutex);
1994         } else if (tfile->napi_enabled) {
1995                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1996                 int queue_len;
1997
1998                 spin_lock_bh(&queue->lock);
1999                 __skb_queue_tail(queue, skb);
2000                 queue_len = skb_queue_len(queue);
2001                 spin_unlock(&queue->lock);
2002
2003                 if (!more || queue_len > NAPI_POLL_WEIGHT)
2004                         napi_schedule(&tfile->napi);
2005
2006                 local_bh_enable();
2007         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
2008                 tun_rx_batched(tun, tfile, skb, more);
2009         } else {
2010                 netif_rx_ni(skb);
2011         }
2012         rcu_read_unlock();
2013
2014         stats = get_cpu_ptr(tun->pcpu_stats);
2015         u64_stats_update_begin(&stats->syncp);
2016         stats->rx_packets++;
2017         stats->rx_bytes += len;
2018         u64_stats_update_end(&stats->syncp);
2019         put_cpu_ptr(stats);
2020
2021         if (rxhash)
2022                 tun_flow_update(tun, rxhash, tfile);
2023
2024         return total_len;
2025 }
2026
2027 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2028 {
2029         struct file *file = iocb->ki_filp;
2030         struct tun_file *tfile = file->private_data;
2031         struct tun_struct *tun = tun_get(tfile);
2032         ssize_t result;
2033
2034         if (!tun)
2035                 return -EBADFD;
2036
2037         result = tun_get_user(tun, tfile, NULL, from,
2038                               file->f_flags & O_NONBLOCK, false);
2039
2040         tun_put(tun);
2041         return result;
2042 }
2043
2044 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2045                                 struct tun_file *tfile,
2046                                 struct xdp_frame *xdp_frame,
2047                                 struct iov_iter *iter)
2048 {
2049         int vnet_hdr_sz = 0;
2050         size_t size = xdp_frame->len;
2051         struct tun_pcpu_stats *stats;
2052         size_t ret;
2053
2054         if (tun->flags & IFF_VNET_HDR) {
2055                 struct virtio_net_hdr gso = { 0 };
2056
2057                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2058                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2059                         return -EINVAL;
2060                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2061                              sizeof(gso)))
2062                         return -EFAULT;
2063                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2064         }
2065
2066         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2067
2068         stats = get_cpu_ptr(tun->pcpu_stats);
2069         u64_stats_update_begin(&stats->syncp);
2070         stats->tx_packets++;
2071         stats->tx_bytes += ret;
2072         u64_stats_update_end(&stats->syncp);
2073         put_cpu_ptr(tun->pcpu_stats);
2074
2075         return ret;
2076 }
2077
2078 /* Put packet to the user space buffer */
2079 static ssize_t tun_put_user(struct tun_struct *tun,
2080                             struct tun_file *tfile,
2081                             struct sk_buff *skb,
2082                             struct iov_iter *iter)
2083 {
2084         struct tun_pi pi = { 0, skb->protocol };
2085         struct tun_pcpu_stats *stats;
2086         ssize_t total;
2087         int vlan_offset = 0;
2088         int vlan_hlen = 0;
2089         int vnet_hdr_sz = 0;
2090
2091         if (skb_vlan_tag_present(skb))
2092                 vlan_hlen = VLAN_HLEN;
2093
2094         if (tun->flags & IFF_VNET_HDR)
2095                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2096
2097         total = skb->len + vlan_hlen + vnet_hdr_sz;
2098
2099         if (!(tun->flags & IFF_NO_PI)) {
2100                 if (iov_iter_count(iter) < sizeof(pi))
2101                         return -EINVAL;
2102
2103                 total += sizeof(pi);
2104                 if (iov_iter_count(iter) < total) {
2105                         /* Packet will be striped */
2106                         pi.flags |= TUN_PKT_STRIP;
2107                 }
2108
2109                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2110                         return -EFAULT;
2111         }
2112
2113         if (vnet_hdr_sz) {
2114                 struct virtio_net_hdr gso;
2115
2116                 if (iov_iter_count(iter) < vnet_hdr_sz)
2117                         return -EINVAL;
2118
2119                 if (virtio_net_hdr_from_skb(skb, &gso,
2120                                             tun_is_little_endian(tun), true,
2121                                             vlan_hlen)) {
2122                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2123                         pr_err("unexpected GSO type: "
2124                                "0x%x, gso_size %d, hdr_len %d\n",
2125                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2126                                tun16_to_cpu(tun, gso.hdr_len));
2127                         print_hex_dump(KERN_ERR, "tun: ",
2128                                        DUMP_PREFIX_NONE,
2129                                        16, 1, skb->head,
2130                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2131                         WARN_ON_ONCE(1);
2132                         return -EINVAL;
2133                 }
2134
2135                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2136                         return -EFAULT;
2137
2138                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2139         }
2140
2141         if (vlan_hlen) {
2142                 int ret;
2143                 struct veth veth;
2144
2145                 veth.h_vlan_proto = skb->vlan_proto;
2146                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2147
2148                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2149
2150                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2151                 if (ret || !iov_iter_count(iter))
2152                         goto done;
2153
2154                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2155                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2156                         goto done;
2157         }
2158
2159         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2160
2161 done:
2162         /* caller is in process context, */
2163         stats = get_cpu_ptr(tun->pcpu_stats);
2164         u64_stats_update_begin(&stats->syncp);
2165         stats->tx_packets++;
2166         stats->tx_bytes += skb->len + vlan_hlen;
2167         u64_stats_update_end(&stats->syncp);
2168         put_cpu_ptr(tun->pcpu_stats);
2169
2170         return total;
2171 }
2172
2173 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2174 {
2175         DECLARE_WAITQUEUE(wait, current);
2176         void *ptr = NULL;
2177         int error = 0;
2178
2179         ptr = ptr_ring_consume(&tfile->tx_ring);
2180         if (ptr)
2181                 goto out;
2182         if (noblock) {
2183                 error = -EAGAIN;
2184                 goto out;
2185         }
2186
2187         add_wait_queue(&tfile->wq.wait, &wait);
2188
2189         while (1) {
2190                 set_current_state(TASK_INTERRUPTIBLE);
2191                 ptr = ptr_ring_consume(&tfile->tx_ring);
2192                 if (ptr)
2193                         break;
2194                 if (signal_pending(current)) {
2195                         error = -ERESTARTSYS;
2196                         break;
2197                 }
2198                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2199                         error = -EFAULT;
2200                         break;
2201                 }
2202
2203                 schedule();
2204         }
2205
2206         __set_current_state(TASK_RUNNING);
2207         remove_wait_queue(&tfile->wq.wait, &wait);
2208
2209 out:
2210         *err = error;
2211         return ptr;
2212 }
2213
2214 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2215                            struct iov_iter *to,
2216                            int noblock, void *ptr)
2217 {
2218         ssize_t ret;
2219         int err;
2220
2221         tun_debug(KERN_INFO, tun, "tun_do_read\n");
2222
2223         if (!iov_iter_count(to)) {
2224                 tun_ptr_free(ptr);
2225                 return 0;
2226         }
2227
2228         if (!ptr) {
2229                 /* Read frames from ring */
2230                 ptr = tun_ring_recv(tfile, noblock, &err);
2231                 if (!ptr)
2232                         return err;
2233         }
2234
2235         if (tun_is_xdp_frame(ptr)) {
2236                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2237
2238                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2239                 xdp_return_frame(xdpf);
2240         } else {
2241                 struct sk_buff *skb = ptr;
2242
2243                 ret = tun_put_user(tun, tfile, skb, to);
2244                 if (unlikely(ret < 0))
2245                         kfree_skb(skb);
2246                 else
2247                         consume_skb(skb);
2248         }
2249
2250         return ret;
2251 }
2252
2253 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2254 {
2255         struct file *file = iocb->ki_filp;
2256         struct tun_file *tfile = file->private_data;
2257         struct tun_struct *tun = tun_get(tfile);
2258         ssize_t len = iov_iter_count(to), ret;
2259
2260         if (!tun)
2261                 return -EBADFD;
2262         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2263         ret = min_t(ssize_t, ret, len);
2264         if (ret > 0)
2265                 iocb->ki_pos = ret;
2266         tun_put(tun);
2267         return ret;
2268 }
2269
2270 static void tun_prog_free(struct rcu_head *rcu)
2271 {
2272         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2273
2274         bpf_prog_destroy(prog->prog);
2275         kfree(prog);
2276 }
2277
2278 static int __tun_set_ebpf(struct tun_struct *tun,
2279                           struct tun_prog __rcu **prog_p,
2280                           struct bpf_prog *prog)
2281 {
2282         struct tun_prog *old, *new = NULL;
2283
2284         if (prog) {
2285                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2286                 if (!new)
2287                         return -ENOMEM;
2288                 new->prog = prog;
2289         }
2290
2291         spin_lock_bh(&tun->lock);
2292         old = rcu_dereference_protected(*prog_p,
2293                                         lockdep_is_held(&tun->lock));
2294         rcu_assign_pointer(*prog_p, new);
2295         spin_unlock_bh(&tun->lock);
2296
2297         if (old)
2298                 call_rcu(&old->rcu, tun_prog_free);
2299
2300         return 0;
2301 }
2302
2303 static void tun_free_netdev(struct net_device *dev)
2304 {
2305         struct tun_struct *tun = netdev_priv(dev);
2306
2307         BUG_ON(!(list_empty(&tun->disabled)));
2308         free_percpu(tun->pcpu_stats);
2309         tun_flow_uninit(tun);
2310         security_tun_dev_free_security(tun->security);
2311         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2312         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2313 }
2314
2315 static void tun_setup(struct net_device *dev)
2316 {
2317         struct tun_struct *tun = netdev_priv(dev);
2318
2319         tun->owner = INVALID_UID;
2320         tun->group = INVALID_GID;
2321         tun_default_link_ksettings(dev, &tun->link_ksettings);
2322
2323         dev->ethtool_ops = &tun_ethtool_ops;
2324         dev->needs_free_netdev = true;
2325         dev->priv_destructor = tun_free_netdev;
2326         /* We prefer our own queue length */
2327         dev->tx_queue_len = TUN_READQ_SIZE;
2328 }
2329
2330 /* Trivial set of netlink ops to allow deleting tun or tap
2331  * device with netlink.
2332  */
2333 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2334                         struct netlink_ext_ack *extack)
2335 {
2336         NL_SET_ERR_MSG(extack,
2337                        "tun/tap creation via rtnetlink is not supported.");
2338         return -EOPNOTSUPP;
2339 }
2340
2341 static size_t tun_get_size(const struct net_device *dev)
2342 {
2343         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2344         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2345
2346         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2347                nla_total_size(sizeof(gid_t)) + /* GROUP */
2348                nla_total_size(sizeof(u8)) + /* TYPE */
2349                nla_total_size(sizeof(u8)) + /* PI */
2350                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2351                nla_total_size(sizeof(u8)) + /* PERSIST */
2352                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2353                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2354                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2355                0;
2356 }
2357
2358 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2359 {
2360         struct tun_struct *tun = netdev_priv(dev);
2361
2362         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2363                 goto nla_put_failure;
2364         if (uid_valid(tun->owner) &&
2365             nla_put_u32(skb, IFLA_TUN_OWNER,
2366                         from_kuid_munged(current_user_ns(), tun->owner)))
2367                 goto nla_put_failure;
2368         if (gid_valid(tun->group) &&
2369             nla_put_u32(skb, IFLA_TUN_GROUP,
2370                         from_kgid_munged(current_user_ns(), tun->group)))
2371                 goto nla_put_failure;
2372         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2373                 goto nla_put_failure;
2374         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2375                 goto nla_put_failure;
2376         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2377                 goto nla_put_failure;
2378         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2379                        !!(tun->flags & IFF_MULTI_QUEUE)))
2380                 goto nla_put_failure;
2381         if (tun->flags & IFF_MULTI_QUEUE) {
2382                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2383                         goto nla_put_failure;
2384                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2385                                 tun->numdisabled))
2386                         goto nla_put_failure;
2387         }
2388
2389         return 0;
2390
2391 nla_put_failure:
2392         return -EMSGSIZE;
2393 }
2394
2395 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2396         .kind           = DRV_NAME,
2397         .priv_size      = sizeof(struct tun_struct),
2398         .setup          = tun_setup,
2399         .validate       = tun_validate,
2400         .get_size       = tun_get_size,
2401         .fill_info      = tun_fill_info,
2402 };
2403
2404 static void tun_sock_write_space(struct sock *sk)
2405 {
2406         struct tun_file *tfile;
2407         wait_queue_head_t *wqueue;
2408
2409         if (!sock_writeable(sk))
2410                 return;
2411
2412         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2413                 return;
2414
2415         wqueue = sk_sleep(sk);
2416         if (wqueue && waitqueue_active(wqueue))
2417                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2418                                                 EPOLLWRNORM | EPOLLWRBAND);
2419
2420         tfile = container_of(sk, struct tun_file, sk);
2421         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2422 }
2423
2424 static void tun_put_page(struct tun_page *tpage)
2425 {
2426         if (tpage->page)
2427                 __page_frag_cache_drain(tpage->page, tpage->count);
2428 }
2429
2430 static int tun_xdp_one(struct tun_struct *tun,
2431                        struct tun_file *tfile,
2432                        struct xdp_buff *xdp, int *flush,
2433                        struct tun_page *tpage)
2434 {
2435         unsigned int datasize = xdp->data_end - xdp->data;
2436         struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2437         struct virtio_net_hdr *gso = &hdr->gso;
2438         struct tun_pcpu_stats *stats;
2439         struct bpf_prog *xdp_prog;
2440         struct sk_buff *skb = NULL;
2441         u32 rxhash = 0, act;
2442         int buflen = hdr->buflen;
2443         int err = 0;
2444         bool skb_xdp = false;
2445         struct page *page;
2446
2447         xdp_prog = rcu_dereference(tun->xdp_prog);
2448         if (xdp_prog) {
2449                 if (gso->gso_type) {
2450                         skb_xdp = true;
2451                         goto build;
2452                 }
2453                 xdp_set_data_meta_invalid(xdp);
2454                 xdp->rxq = &tfile->xdp_rxq;
2455
2456                 act = bpf_prog_run_xdp(xdp_prog, xdp);
2457                 err = tun_xdp_act(tun, xdp_prog, xdp, act);
2458                 if (err < 0) {
2459                         put_page(virt_to_head_page(xdp->data));
2460                         return err;
2461                 }
2462
2463                 switch (err) {
2464                 case XDP_REDIRECT:
2465                         *flush = true;
2466                         /* fall through */
2467                 case XDP_TX:
2468                         return 0;
2469                 case XDP_PASS:
2470                         break;
2471                 default:
2472                         page = virt_to_head_page(xdp->data);
2473                         if (tpage->page == page) {
2474                                 ++tpage->count;
2475                         } else {
2476                                 tun_put_page(tpage);
2477                                 tpage->page = page;
2478                                 tpage->count = 1;
2479                         }
2480                         return 0;
2481                 }
2482         }
2483
2484 build:
2485         skb = build_skb(xdp->data_hard_start, buflen);
2486         if (!skb) {
2487                 err = -ENOMEM;
2488                 goto out;
2489         }
2490
2491         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2492         skb_put(skb, xdp->data_end - xdp->data);
2493
2494         if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2495                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
2496                 kfree_skb(skb);
2497                 err = -EINVAL;
2498                 goto out;
2499         }
2500
2501         skb->protocol = eth_type_trans(skb, tun->dev);
2502         skb_reset_network_header(skb);
2503         skb_probe_transport_header(skb);
2504
2505         if (skb_xdp) {
2506                 err = do_xdp_generic(xdp_prog, skb);
2507                 if (err != XDP_PASS)
2508                         goto out;
2509         }
2510
2511         if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2512             !tfile->detached)
2513                 rxhash = __skb_get_hash_symmetric(skb);
2514
2515         skb_record_rx_queue(skb, tfile->queue_index);
2516         netif_receive_skb(skb);
2517
2518         /* No need for get_cpu_ptr() here since this function is
2519          * always called with bh disabled
2520          */
2521         stats = this_cpu_ptr(tun->pcpu_stats);
2522         u64_stats_update_begin(&stats->syncp);
2523         stats->rx_packets++;
2524         stats->rx_bytes += datasize;
2525         u64_stats_update_end(&stats->syncp);
2526
2527         if (rxhash)
2528                 tun_flow_update(tun, rxhash, tfile);
2529
2530 out:
2531         return err;
2532 }
2533
2534 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2535 {
2536         int ret, i;
2537         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2538         struct tun_struct *tun = tun_get(tfile);
2539         struct tun_msg_ctl *ctl = m->msg_control;
2540         struct xdp_buff *xdp;
2541
2542         if (!tun)
2543                 return -EBADFD;
2544
2545         if (ctl && (ctl->type == TUN_MSG_PTR)) {
2546                 struct tun_page tpage;
2547                 int n = ctl->num;
2548                 int flush = 0;
2549
2550                 memset(&tpage, 0, sizeof(tpage));
2551
2552                 local_bh_disable();
2553                 rcu_read_lock();
2554
2555                 for (i = 0; i < n; i++) {
2556                         xdp = &((struct xdp_buff *)ctl->ptr)[i];
2557                         tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2558                 }
2559
2560                 if (flush)
2561                         xdp_do_flush_map();
2562
2563                 rcu_read_unlock();
2564                 local_bh_enable();
2565
2566                 tun_put_page(&tpage);
2567
2568                 ret = total_len;
2569                 goto out;
2570         }
2571
2572         ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2573                            m->msg_flags & MSG_DONTWAIT,
2574                            m->msg_flags & MSG_MORE);
2575 out:
2576         tun_put(tun);
2577         return ret;
2578 }
2579
2580 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2581                        int flags)
2582 {
2583         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2584         struct tun_struct *tun = tun_get(tfile);
2585         void *ptr = m->msg_control;
2586         int ret;
2587
2588         if (!tun) {
2589                 ret = -EBADFD;
2590                 goto out_free;
2591         }
2592
2593         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2594                 ret = -EINVAL;
2595                 goto out_put_tun;
2596         }
2597         if (flags & MSG_ERRQUEUE) {
2598                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2599                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2600                 goto out;
2601         }
2602         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2603         if (ret > (ssize_t)total_len) {
2604                 m->msg_flags |= MSG_TRUNC;
2605                 ret = flags & MSG_TRUNC ? ret : total_len;
2606         }
2607 out:
2608         tun_put(tun);
2609         return ret;
2610
2611 out_put_tun:
2612         tun_put(tun);
2613 out_free:
2614         tun_ptr_free(ptr);
2615         return ret;
2616 }
2617
2618 static int tun_ptr_peek_len(void *ptr)
2619 {
2620         if (likely(ptr)) {
2621                 if (tun_is_xdp_frame(ptr)) {
2622                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2623
2624                         return xdpf->len;
2625                 }
2626                 return __skb_array_len_with_tag(ptr);
2627         } else {
2628                 return 0;
2629         }
2630 }
2631
2632 static int tun_peek_len(struct socket *sock)
2633 {
2634         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2635         struct tun_struct *tun;
2636         int ret = 0;
2637
2638         tun = tun_get(tfile);
2639         if (!tun)
2640                 return 0;
2641
2642         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2643         tun_put(tun);
2644
2645         return ret;
2646 }
2647
2648 /* Ops structure to mimic raw sockets with tun */
2649 static const struct proto_ops tun_socket_ops = {
2650         .peek_len = tun_peek_len,
2651         .sendmsg = tun_sendmsg,
2652         .recvmsg = tun_recvmsg,
2653 };
2654
2655 static struct proto tun_proto = {
2656         .name           = "tun",
2657         .owner          = THIS_MODULE,
2658         .obj_size       = sizeof(struct tun_file),
2659 };
2660
2661 static int tun_flags(struct tun_struct *tun)
2662 {
2663         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2664 }
2665
2666 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2667                               char *buf)
2668 {
2669         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2670         return sprintf(buf, "0x%x\n", tun_flags(tun));
2671 }
2672
2673 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2674                               char *buf)
2675 {
2676         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2677         return uid_valid(tun->owner)?
2678                 sprintf(buf, "%u\n",
2679                         from_kuid_munged(current_user_ns(), tun->owner)):
2680                 sprintf(buf, "-1\n");
2681 }
2682
2683 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2684                               char *buf)
2685 {
2686         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2687         return gid_valid(tun->group) ?
2688                 sprintf(buf, "%u\n",
2689                         from_kgid_munged(current_user_ns(), tun->group)):
2690                 sprintf(buf, "-1\n");
2691 }
2692
2693 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2694 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2695 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2696
2697 static struct attribute *tun_dev_attrs[] = {
2698         &dev_attr_tun_flags.attr,
2699         &dev_attr_owner.attr,
2700         &dev_attr_group.attr,
2701         NULL
2702 };
2703
2704 static const struct attribute_group tun_attr_group = {
2705         .attrs = tun_dev_attrs
2706 };
2707
2708 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2709 {
2710         struct tun_struct *tun;
2711         struct tun_file *tfile = file->private_data;
2712         struct net_device *dev;
2713         int err;
2714
2715         if (tfile->detached)
2716                 return -EINVAL;
2717
2718         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2719                 if (!capable(CAP_NET_ADMIN))
2720                         return -EPERM;
2721
2722                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2723                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2724                         return -EINVAL;
2725         }
2726
2727         dev = __dev_get_by_name(net, ifr->ifr_name);
2728         if (dev) {
2729                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2730                         return -EBUSY;
2731                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2732                         tun = netdev_priv(dev);
2733                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2734                         tun = netdev_priv(dev);
2735                 else
2736                         return -EINVAL;
2737
2738                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2739                     !!(tun->flags & IFF_MULTI_QUEUE))
2740                         return -EINVAL;
2741
2742                 if (tun_not_capable(tun))
2743                         return -EPERM;
2744                 err = security_tun_dev_open(tun->security);
2745                 if (err < 0)
2746                         return err;
2747
2748                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2749                                  ifr->ifr_flags & IFF_NAPI,
2750                                  ifr->ifr_flags & IFF_NAPI_FRAGS);
2751                 if (err < 0)
2752                         return err;
2753
2754                 if (tun->flags & IFF_MULTI_QUEUE &&
2755                     (tun->numqueues + tun->numdisabled > 1)) {
2756                         /* One or more queue has already been attached, no need
2757                          * to initialize the device again.
2758                          */
2759                         netdev_state_change(dev);
2760                         return 0;
2761                 }
2762
2763                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2764                               (ifr->ifr_flags & TUN_FEATURES);
2765
2766                 netdev_state_change(dev);
2767         } else {
2768                 char *name;
2769                 unsigned long flags = 0;
2770                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2771                              MAX_TAP_QUEUES : 1;
2772
2773                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2774                         return -EPERM;
2775                 err = security_tun_dev_create();
2776                 if (err < 0)
2777                         return err;
2778
2779                 /* Set dev type */
2780                 if (ifr->ifr_flags & IFF_TUN) {
2781                         /* TUN device */
2782                         flags |= IFF_TUN;
2783                         name = "tun%d";
2784                 } else if (ifr->ifr_flags & IFF_TAP) {
2785                         /* TAP device */
2786                         flags |= IFF_TAP;
2787                         name = "tap%d";
2788                 } else
2789                         return -EINVAL;
2790
2791                 if (*ifr->ifr_name)
2792                         name = ifr->ifr_name;
2793
2794                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2795                                        NET_NAME_UNKNOWN, tun_setup, queues,
2796                                        queues);
2797
2798                 if (!dev)
2799                         return -ENOMEM;
2800                 err = dev_get_valid_name(net, dev, name);
2801                 if (err < 0)
2802                         goto err_free_dev;
2803
2804                 dev_net_set(dev, net);
2805                 dev->rtnl_link_ops = &tun_link_ops;
2806                 dev->ifindex = tfile->ifindex;
2807                 dev->sysfs_groups[0] = &tun_attr_group;
2808
2809                 tun = netdev_priv(dev);
2810                 tun->dev = dev;
2811                 tun->flags = flags;
2812                 tun->txflt.count = 0;
2813                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2814
2815                 tun->align = NET_SKB_PAD;
2816                 tun->filter_attached = false;
2817                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2818                 tun->rx_batched = 0;
2819                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2820
2821                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2822                 if (!tun->pcpu_stats) {
2823                         err = -ENOMEM;
2824                         goto err_free_dev;
2825                 }
2826
2827                 spin_lock_init(&tun->lock);
2828
2829                 err = security_tun_dev_alloc_security(&tun->security);
2830                 if (err < 0)
2831                         goto err_free_stat;
2832
2833                 tun_net_init(dev);
2834                 tun_flow_init(tun);
2835
2836                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2837                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2838                                    NETIF_F_HW_VLAN_STAG_TX;
2839                 dev->features = dev->hw_features | NETIF_F_LLTX;
2840                 dev->vlan_features = dev->features &
2841                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2842                                        NETIF_F_HW_VLAN_STAG_TX);
2843
2844                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2845                               (ifr->ifr_flags & TUN_FEATURES);
2846
2847                 INIT_LIST_HEAD(&tun->disabled);
2848                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI,
2849                                  ifr->ifr_flags & IFF_NAPI_FRAGS);
2850                 if (err < 0)
2851                         goto err_free_flow;
2852
2853                 err = register_netdevice(tun->dev);
2854                 if (err < 0)
2855                         goto err_detach;
2856         }
2857
2858         netif_carrier_on(tun->dev);
2859
2860         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2861
2862         /* Make sure persistent devices do not get stuck in
2863          * xoff state.
2864          */
2865         if (netif_running(tun->dev))
2866                 netif_tx_wake_all_queues(tun->dev);
2867
2868         strcpy(ifr->ifr_name, tun->dev->name);
2869         return 0;
2870
2871 err_detach:
2872         tun_detach_all(dev);
2873         /* register_netdevice() already called tun_free_netdev() */
2874         goto err_free_dev;
2875
2876 err_free_flow:
2877         tun_flow_uninit(tun);
2878         security_tun_dev_free_security(tun->security);
2879 err_free_stat:
2880         free_percpu(tun->pcpu_stats);
2881 err_free_dev:
2882         free_netdev(dev);
2883         return err;
2884 }
2885
2886 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2887 {
2888         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2889
2890         strcpy(ifr->ifr_name, tun->dev->name);
2891
2892         ifr->ifr_flags = tun_flags(tun);
2893
2894 }
2895
2896 /* This is like a cut-down ethtool ops, except done via tun fd so no
2897  * privs required. */
2898 static int set_offload(struct tun_struct *tun, unsigned long arg)
2899 {
2900         netdev_features_t features = 0;
2901
2902         if (arg & TUN_F_CSUM) {
2903                 features |= NETIF_F_HW_CSUM;
2904                 arg &= ~TUN_F_CSUM;
2905
2906                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2907                         if (arg & TUN_F_TSO_ECN) {
2908                                 features |= NETIF_F_TSO_ECN;
2909                                 arg &= ~TUN_F_TSO_ECN;
2910                         }
2911                         if (arg & TUN_F_TSO4)
2912                                 features |= NETIF_F_TSO;
2913                         if (arg & TUN_F_TSO6)
2914                                 features |= NETIF_F_TSO6;
2915                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2916                 }
2917
2918                 arg &= ~TUN_F_UFO;
2919         }
2920
2921         /* This gives the user a way to test for new features in future by
2922          * trying to set them. */
2923         if (arg)
2924                 return -EINVAL;
2925
2926         tun->set_features = features;
2927         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2928         tun->dev->wanted_features |= features;
2929         netdev_update_features(tun->dev);
2930
2931         return 0;
2932 }
2933
2934 static void tun_detach_filter(struct tun_struct *tun, int n)
2935 {
2936         int i;
2937         struct tun_file *tfile;
2938
2939         for (i = 0; i < n; i++) {
2940                 tfile = rtnl_dereference(tun->tfiles[i]);
2941                 lock_sock(tfile->socket.sk);
2942                 sk_detach_filter(tfile->socket.sk);
2943                 release_sock(tfile->socket.sk);
2944         }
2945
2946         tun->filter_attached = false;
2947 }
2948
2949 static int tun_attach_filter(struct tun_struct *tun)
2950 {
2951         int i, ret = 0;
2952         struct tun_file *tfile;
2953
2954         for (i = 0; i < tun->numqueues; i++) {
2955                 tfile = rtnl_dereference(tun->tfiles[i]);
2956                 lock_sock(tfile->socket.sk);
2957                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2958                 release_sock(tfile->socket.sk);
2959                 if (ret) {
2960                         tun_detach_filter(tun, i);
2961                         return ret;
2962                 }
2963         }
2964
2965         tun->filter_attached = true;
2966         return ret;
2967 }
2968
2969 static void tun_set_sndbuf(struct tun_struct *tun)
2970 {
2971         struct tun_file *tfile;
2972         int i;
2973
2974         for (i = 0; i < tun->numqueues; i++) {
2975                 tfile = rtnl_dereference(tun->tfiles[i]);
2976                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2977         }
2978 }
2979
2980 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2981 {
2982         struct tun_file *tfile = file->private_data;
2983         struct tun_struct *tun;
2984         int ret = 0;
2985
2986         rtnl_lock();
2987
2988         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2989                 tun = tfile->detached;
2990                 if (!tun) {
2991                         ret = -EINVAL;
2992                         goto unlock;
2993                 }
2994                 ret = security_tun_dev_attach_queue(tun->security);
2995                 if (ret < 0)
2996                         goto unlock;
2997                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2998                                  tun->flags & IFF_NAPI_FRAGS);
2999         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
3000                 tun = rtnl_dereference(tfile->tun);
3001                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
3002                         ret = -EINVAL;
3003                 else
3004                         __tun_detach(tfile, false);
3005         } else
3006                 ret = -EINVAL;
3007
3008         if (ret >= 0)
3009                 netdev_state_change(tun->dev);
3010
3011 unlock:
3012         rtnl_unlock();
3013         return ret;
3014 }
3015
3016 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
3017                         void __user *data)
3018 {
3019         struct bpf_prog *prog;
3020         int fd;
3021
3022         if (copy_from_user(&fd, data, sizeof(fd)))
3023                 return -EFAULT;
3024
3025         if (fd == -1) {
3026                 prog = NULL;
3027         } else {
3028                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
3029                 if (IS_ERR(prog))
3030                         return PTR_ERR(prog);
3031         }
3032
3033         return __tun_set_ebpf(tun, prog_p, prog);
3034 }
3035
3036 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
3037                             unsigned long arg, int ifreq_len)
3038 {
3039         struct tun_file *tfile = file->private_data;
3040         struct net *net = sock_net(&tfile->sk);
3041         struct tun_struct *tun;
3042         void __user* argp = (void __user*)arg;
3043         unsigned int ifindex, carrier;
3044         struct ifreq ifr;
3045         kuid_t owner;
3046         kgid_t group;
3047         int sndbuf;
3048         int vnet_hdr_sz;
3049         int le;
3050         int ret;
3051         bool do_notify = false;
3052
3053         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3054             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3055                 if (copy_from_user(&ifr, argp, ifreq_len))
3056                         return -EFAULT;
3057         } else {
3058                 memset(&ifr, 0, sizeof(ifr));
3059         }
3060         if (cmd == TUNGETFEATURES) {
3061                 /* Currently this just means: "what IFF flags are valid?".
3062                  * This is needed because we never checked for invalid flags on
3063                  * TUNSETIFF.
3064                  */
3065                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
3066                                 (unsigned int __user*)argp);
3067         } else if (cmd == TUNSETQUEUE) {
3068                 return tun_set_queue(file, &ifr);
3069         } else if (cmd == SIOCGSKNS) {
3070                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3071                         return -EPERM;
3072                 return open_related_ns(&net->ns, get_net_ns);
3073         }
3074
3075         ret = 0;
3076         rtnl_lock();
3077
3078         tun = tun_get(tfile);
3079         if (cmd == TUNSETIFF) {
3080                 ret = -EEXIST;
3081                 if (tun)
3082                         goto unlock;
3083
3084                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3085
3086                 ret = tun_set_iff(net, file, &ifr);
3087
3088                 if (ret)
3089                         goto unlock;
3090
3091                 if (copy_to_user(argp, &ifr, ifreq_len))
3092                         ret = -EFAULT;
3093                 goto unlock;
3094         }
3095         if (cmd == TUNSETIFINDEX) {
3096                 ret = -EPERM;
3097                 if (tun)
3098                         goto unlock;
3099
3100                 ret = -EFAULT;
3101                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3102                         goto unlock;
3103
3104                 ret = 0;
3105                 tfile->ifindex = ifindex;
3106                 goto unlock;
3107         }
3108
3109         ret = -EBADFD;
3110         if (!tun)
3111                 goto unlock;
3112
3113         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
3114
3115         net = dev_net(tun->dev);
3116         ret = 0;
3117         switch (cmd) {
3118         case TUNGETIFF:
3119                 tun_get_iff(tun, &ifr);
3120
3121                 if (tfile->detached)
3122                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
3123                 if (!tfile->socket.sk->sk_filter)
3124                         ifr.ifr_flags |= IFF_NOFILTER;
3125
3126                 if (copy_to_user(argp, &ifr, ifreq_len))
3127                         ret = -EFAULT;
3128                 break;
3129
3130         case TUNSETNOCSUM:
3131                 /* Disable/Enable checksum */
3132
3133                 /* [unimplemented] */
3134                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
3135                           arg ? "disabled" : "enabled");
3136                 break;
3137
3138         case TUNSETPERSIST:
3139                 /* Disable/Enable persist mode. Keep an extra reference to the
3140                  * module to prevent the module being unprobed.
3141                  */
3142                 if (arg && !(tun->flags & IFF_PERSIST)) {
3143                         tun->flags |= IFF_PERSIST;
3144                         __module_get(THIS_MODULE);
3145                         do_notify = true;
3146                 }
3147                 if (!arg && (tun->flags & IFF_PERSIST)) {
3148                         tun->flags &= ~IFF_PERSIST;
3149                         module_put(THIS_MODULE);
3150                         do_notify = true;
3151                 }
3152
3153                 tun_debug(KERN_INFO, tun, "persist %s\n",
3154                           arg ? "enabled" : "disabled");
3155                 break;
3156
3157         case TUNSETOWNER:
3158                 /* Set owner of the device */
3159                 owner = make_kuid(current_user_ns(), arg);
3160                 if (!uid_valid(owner)) {
3161                         ret = -EINVAL;
3162                         break;
3163                 }
3164                 tun->owner = owner;
3165                 do_notify = true;
3166                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
3167                           from_kuid(&init_user_ns, tun->owner));
3168                 break;
3169
3170         case TUNSETGROUP:
3171                 /* Set group of the device */
3172                 group = make_kgid(current_user_ns(), arg);
3173                 if (!gid_valid(group)) {
3174                         ret = -EINVAL;
3175                         break;
3176                 }
3177                 tun->group = group;
3178                 do_notify = true;
3179                 tun_debug(KERN_INFO, tun, "group set to %u\n",
3180                           from_kgid(&init_user_ns, tun->group));
3181                 break;
3182
3183         case TUNSETLINK:
3184                 /* Only allow setting the type when the interface is down */
3185                 if (tun->dev->flags & IFF_UP) {
3186                         tun_debug(KERN_INFO, tun,
3187                                   "Linktype set failed because interface is up\n");
3188                         ret = -EBUSY;
3189                 } else {
3190                         tun->dev->type = (int) arg;
3191                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
3192                                   tun->dev->type);
3193                         ret = 0;
3194                 }
3195                 break;
3196
3197 #ifdef TUN_DEBUG
3198         case TUNSETDEBUG:
3199                 tun->debug = arg;
3200                 break;
3201 #endif
3202         case TUNSETOFFLOAD:
3203                 ret = set_offload(tun, arg);
3204                 break;
3205
3206         case TUNSETTXFILTER:
3207                 /* Can be set only for TAPs */
3208                 ret = -EINVAL;
3209                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3210                         break;
3211                 ret = update_filter(&tun->txflt, (void __user *)arg);
3212                 break;
3213
3214         case SIOCGIFHWADDR:
3215                 /* Get hw address */
3216                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3217                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3218                 if (copy_to_user(argp, &ifr, ifreq_len))
3219                         ret = -EFAULT;
3220                 break;
3221
3222         case SIOCSIFHWADDR:
3223                 /* Set hw address */
3224                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3225                           ifr.ifr_hwaddr.sa_data);
3226
3227                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr, NULL);
3228                 break;
3229
3230         case TUNGETSNDBUF:
3231                 sndbuf = tfile->socket.sk->sk_sndbuf;
3232                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3233                         ret = -EFAULT;
3234                 break;
3235
3236         case TUNSETSNDBUF:
3237                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3238                         ret = -EFAULT;
3239                         break;
3240                 }
3241                 if (sndbuf <= 0) {
3242                         ret = -EINVAL;
3243                         break;
3244                 }
3245
3246                 tun->sndbuf = sndbuf;
3247                 tun_set_sndbuf(tun);
3248                 break;
3249
3250         case TUNGETVNETHDRSZ:
3251                 vnet_hdr_sz = tun->vnet_hdr_sz;
3252                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3253                         ret = -EFAULT;
3254                 break;
3255
3256         case TUNSETVNETHDRSZ:
3257                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3258                         ret = -EFAULT;
3259                         break;
3260                 }
3261                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3262                         ret = -EINVAL;
3263                         break;
3264                 }
3265
3266                 tun->vnet_hdr_sz = vnet_hdr_sz;
3267                 break;
3268
3269         case TUNGETVNETLE:
3270                 le = !!(tun->flags & TUN_VNET_LE);
3271                 if (put_user(le, (int __user *)argp))
3272                         ret = -EFAULT;
3273                 break;
3274
3275         case TUNSETVNETLE:
3276                 if (get_user(le, (int __user *)argp)) {
3277                         ret = -EFAULT;
3278                         break;
3279                 }
3280                 if (le)
3281                         tun->flags |= TUN_VNET_LE;
3282                 else
3283                         tun->flags &= ~TUN_VNET_LE;
3284                 break;
3285
3286         case TUNGETVNETBE:
3287                 ret = tun_get_vnet_be(tun, argp);
3288                 break;
3289
3290         case TUNSETVNETBE:
3291                 ret = tun_set_vnet_be(tun, argp);
3292                 break;
3293
3294         case TUNATTACHFILTER:
3295                 /* Can be set only for TAPs */
3296                 ret = -EINVAL;
3297                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3298                         break;
3299                 ret = -EFAULT;
3300                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3301                         break;
3302
3303                 ret = tun_attach_filter(tun);
3304                 break;
3305
3306         case TUNDETACHFILTER:
3307                 /* Can be set only for TAPs */
3308                 ret = -EINVAL;
3309                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3310                         break;
3311                 ret = 0;
3312                 tun_detach_filter(tun, tun->numqueues);
3313                 break;
3314
3315         case TUNGETFILTER:
3316                 ret = -EINVAL;
3317                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3318                         break;
3319                 ret = -EFAULT;
3320                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3321                         break;
3322                 ret = 0;
3323                 break;
3324
3325         case TUNSETSTEERINGEBPF:
3326                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3327                 break;
3328
3329         case TUNSETFILTEREBPF:
3330                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3331                 break;
3332
3333         case TUNSETCARRIER:
3334                 ret = -EFAULT;
3335                 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3336                         goto unlock;
3337
3338                 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3339                 break;
3340
3341         case TUNGETDEVNETNS:
3342                 ret = -EPERM;
3343                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3344                         goto unlock;
3345                 ret = open_related_ns(&net->ns, get_net_ns);
3346                 break;
3347
3348         default:
3349                 ret = -EINVAL;
3350                 break;
3351         }
3352
3353         if (do_notify)
3354                 netdev_state_change(tun->dev);
3355
3356 unlock:
3357         rtnl_unlock();
3358         if (tun)
3359                 tun_put(tun);
3360         return ret;
3361 }
3362
3363 static long tun_chr_ioctl(struct file *file,
3364                           unsigned int cmd, unsigned long arg)
3365 {
3366         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3367 }
3368
3369 #ifdef CONFIG_COMPAT
3370 static long tun_chr_compat_ioctl(struct file *file,
3371                          unsigned int cmd, unsigned long arg)
3372 {
3373         switch (cmd) {
3374         case TUNSETIFF:
3375         case TUNGETIFF:
3376         case TUNSETTXFILTER:
3377         case TUNGETSNDBUF:
3378         case TUNSETSNDBUF:
3379         case SIOCGIFHWADDR:
3380         case SIOCSIFHWADDR:
3381                 arg = (unsigned long)compat_ptr(arg);
3382                 break;
3383         default:
3384                 arg = (compat_ulong_t)arg;
3385                 break;
3386         }
3387
3388         /*
3389          * compat_ifreq is shorter than ifreq, so we must not access beyond
3390          * the end of that structure. All fields that are used in this
3391          * driver are compatible though, we don't need to convert the
3392          * contents.
3393          */
3394         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3395 }
3396 #endif /* CONFIG_COMPAT */
3397
3398 static int tun_chr_fasync(int fd, struct file *file, int on)
3399 {
3400         struct tun_file *tfile = file->private_data;
3401         int ret;
3402
3403         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3404                 goto out;
3405
3406         if (on) {
3407                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3408                 tfile->flags |= TUN_FASYNC;
3409         } else
3410                 tfile->flags &= ~TUN_FASYNC;
3411         ret = 0;
3412 out:
3413         return ret;
3414 }
3415
3416 static int tun_chr_open(struct inode *inode, struct file * file)
3417 {
3418         struct net *net = current->nsproxy->net_ns;
3419         struct tun_file *tfile;
3420
3421         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3422
3423         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3424                                             &tun_proto, 0);
3425         if (!tfile)
3426                 return -ENOMEM;
3427         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3428                 sk_free(&tfile->sk);
3429                 return -ENOMEM;
3430         }
3431
3432         mutex_init(&tfile->napi_mutex);
3433         RCU_INIT_POINTER(tfile->tun, NULL);
3434         tfile->flags = 0;
3435         tfile->ifindex = 0;
3436
3437         init_waitqueue_head(&tfile->wq.wait);
3438         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3439
3440         tfile->socket.file = file;
3441         tfile->socket.ops = &tun_socket_ops;
3442
3443         sock_init_data(&tfile->socket, &tfile->sk);
3444
3445         tfile->sk.sk_write_space = tun_sock_write_space;
3446         tfile->sk.sk_sndbuf = INT_MAX;
3447
3448         file->private_data = tfile;
3449         INIT_LIST_HEAD(&tfile->next);
3450
3451         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3452
3453         return 0;
3454 }
3455
3456 static int tun_chr_close(struct inode *inode, struct file *file)
3457 {
3458         struct tun_file *tfile = file->private_data;
3459
3460         tun_detach(tfile, true);
3461
3462         return 0;
3463 }
3464
3465 #ifdef CONFIG_PROC_FS
3466 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3467 {
3468         struct tun_file *tfile = file->private_data;
3469         struct tun_struct *tun;
3470         struct ifreq ifr;
3471
3472         memset(&ifr, 0, sizeof(ifr));
3473
3474         rtnl_lock();
3475         tun = tun_get(tfile);
3476         if (tun)
3477                 tun_get_iff(tun, &ifr);
3478         rtnl_unlock();
3479
3480         if (tun)
3481                 tun_put(tun);
3482
3483         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3484 }
3485 #endif
3486
3487 static const struct file_operations tun_fops = {
3488         .owner  = THIS_MODULE,
3489         .llseek = no_llseek,
3490         .read_iter  = tun_chr_read_iter,
3491         .write_iter = tun_chr_write_iter,
3492         .poll   = tun_chr_poll,
3493         .unlocked_ioctl = tun_chr_ioctl,
3494 #ifdef CONFIG_COMPAT
3495         .compat_ioctl = tun_chr_compat_ioctl,
3496 #endif
3497         .open   = tun_chr_open,
3498         .release = tun_chr_close,
3499         .fasync = tun_chr_fasync,
3500 #ifdef CONFIG_PROC_FS
3501         .show_fdinfo = tun_chr_show_fdinfo,
3502 #endif
3503 };
3504
3505 static struct miscdevice tun_miscdev = {
3506         .minor = TUN_MINOR,
3507         .name = "tun",
3508         .nodename = "net/tun",
3509         .fops = &tun_fops,
3510 };
3511
3512 /* ethtool interface */
3513
3514 static void tun_default_link_ksettings(struct net_device *dev,
3515                                        struct ethtool_link_ksettings *cmd)
3516 {
3517         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3518         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3519         cmd->base.speed         = SPEED_10;
3520         cmd->base.duplex        = DUPLEX_FULL;
3521         cmd->base.port          = PORT_TP;
3522         cmd->base.phy_address   = 0;
3523         cmd->base.autoneg       = AUTONEG_DISABLE;
3524 }
3525
3526 static int tun_get_link_ksettings(struct net_device *dev,
3527                                   struct ethtool_link_ksettings *cmd)
3528 {
3529         struct tun_struct *tun = netdev_priv(dev);
3530
3531         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3532         return 0;
3533 }
3534
3535 static int tun_set_link_ksettings(struct net_device *dev,
3536                                   const struct ethtool_link_ksettings *cmd)
3537 {
3538         struct tun_struct *tun = netdev_priv(dev);
3539
3540         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3541         return 0;
3542 }
3543
3544 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3545 {
3546         struct tun_struct *tun = netdev_priv(dev);
3547
3548         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3549         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3550
3551         switch (tun->flags & TUN_TYPE_MASK) {
3552         case IFF_TUN:
3553                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3554                 break;
3555         case IFF_TAP:
3556                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3557                 break;
3558         }
3559 }
3560
3561 static u32 tun_get_msglevel(struct net_device *dev)
3562 {
3563 #ifdef TUN_DEBUG
3564         struct tun_struct *tun = netdev_priv(dev);
3565         return tun->debug;
3566 #else
3567         return -EOPNOTSUPP;
3568 #endif
3569 }
3570
3571 static void tun_set_msglevel(struct net_device *dev, u32 value)
3572 {
3573 #ifdef TUN_DEBUG
3574         struct tun_struct *tun = netdev_priv(dev);
3575         tun->debug = value;
3576 #endif
3577 }
3578
3579 static int tun_get_coalesce(struct net_device *dev,
3580                             struct ethtool_coalesce *ec)
3581 {
3582         struct tun_struct *tun = netdev_priv(dev);
3583
3584         ec->rx_max_coalesced_frames = tun->rx_batched;
3585
3586         return 0;
3587 }
3588
3589 static int tun_set_coalesce(struct net_device *dev,
3590                             struct ethtool_coalesce *ec)
3591 {
3592         struct tun_struct *tun = netdev_priv(dev);
3593
3594         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3595                 tun->rx_batched = NAPI_POLL_WEIGHT;
3596         else
3597                 tun->rx_batched = ec->rx_max_coalesced_frames;
3598
3599         return 0;
3600 }
3601
3602 static const struct ethtool_ops tun_ethtool_ops = {
3603         .get_drvinfo    = tun_get_drvinfo,
3604         .get_msglevel   = tun_get_msglevel,
3605         .set_msglevel   = tun_set_msglevel,
3606         .get_link       = ethtool_op_get_link,
3607         .get_ts_info    = ethtool_op_get_ts_info,
3608         .get_coalesce   = tun_get_coalesce,
3609         .set_coalesce   = tun_set_coalesce,
3610         .get_link_ksettings = tun_get_link_ksettings,
3611         .set_link_ksettings = tun_set_link_ksettings,
3612 };
3613
3614 static int tun_queue_resize(struct tun_struct *tun)
3615 {
3616         struct net_device *dev = tun->dev;
3617         struct tun_file *tfile;
3618         struct ptr_ring **rings;
3619         int n = tun->numqueues + tun->numdisabled;
3620         int ret, i;
3621
3622         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3623         if (!rings)
3624                 return -ENOMEM;
3625
3626         for (i = 0; i < tun->numqueues; i++) {
3627                 tfile = rtnl_dereference(tun->tfiles[i]);
3628                 rings[i] = &tfile->tx_ring;
3629         }
3630         list_for_each_entry(tfile, &tun->disabled, next)
3631                 rings[i++] = &tfile->tx_ring;
3632
3633         ret = ptr_ring_resize_multiple(rings, n,
3634                                        dev->tx_queue_len, GFP_KERNEL,
3635                                        tun_ptr_free);
3636
3637         kfree(rings);
3638         return ret;
3639 }
3640
3641 static int tun_device_event(struct notifier_block *unused,
3642                             unsigned long event, void *ptr)
3643 {
3644         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3645         struct tun_struct *tun = netdev_priv(dev);
3646
3647         if (dev->rtnl_link_ops != &tun_link_ops)
3648                 return NOTIFY_DONE;
3649
3650         switch (event) {
3651         case NETDEV_CHANGE_TX_QUEUE_LEN:
3652                 if (tun_queue_resize(tun))
3653                         return NOTIFY_BAD;
3654                 break;
3655         default:
3656                 break;
3657         }
3658
3659         return NOTIFY_DONE;
3660 }
3661
3662 static struct notifier_block tun_notifier_block __read_mostly = {
3663         .notifier_call  = tun_device_event,
3664 };
3665
3666 static int __init tun_init(void)
3667 {
3668         int ret = 0;
3669
3670         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3671
3672         ret = rtnl_link_register(&tun_link_ops);
3673         if (ret) {
3674                 pr_err("Can't register link_ops\n");
3675                 goto err_linkops;
3676         }
3677
3678         ret = misc_register(&tun_miscdev);
3679         if (ret) {
3680                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3681                 goto err_misc;
3682         }
3683
3684         ret = register_netdevice_notifier(&tun_notifier_block);
3685         if (ret) {
3686                 pr_err("Can't register netdevice notifier\n");
3687                 goto err_notifier;
3688         }
3689
3690         return  0;
3691
3692 err_notifier:
3693         misc_deregister(&tun_miscdev);
3694 err_misc:
3695         rtnl_link_unregister(&tun_link_ops);
3696 err_linkops:
3697         return ret;
3698 }
3699
3700 static void tun_cleanup(void)
3701 {
3702         misc_deregister(&tun_miscdev);
3703         rtnl_link_unregister(&tun_link_ops);
3704         unregister_netdevice_notifier(&tun_notifier_block);
3705 }
3706
3707 /* Get an underlying socket object from tun file.  Returns error unless file is
3708  * attached to a device.  The returned object works like a packet socket, it
3709  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3710  * holding a reference to the file for as long as the socket is in use. */
3711 struct socket *tun_get_socket(struct file *file)
3712 {
3713         struct tun_file *tfile;
3714         if (file->f_op != &tun_fops)
3715                 return ERR_PTR(-EINVAL);
3716         tfile = file->private_data;
3717         if (!tfile)
3718                 return ERR_PTR(-EBADFD);
3719         return &tfile->socket;
3720 }
3721 EXPORT_SYMBOL_GPL(tun_get_socket);
3722
3723 struct ptr_ring *tun_get_tx_ring(struct file *file)
3724 {
3725         struct tun_file *tfile;
3726
3727         if (file->f_op != &tun_fops)
3728                 return ERR_PTR(-EINVAL);
3729         tfile = file->private_data;
3730         if (!tfile)
3731                 return ERR_PTR(-EBADFD);
3732         return &tfile->tx_ring;
3733 }
3734 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3735
3736 module_init(tun_init);
3737 module_exit(tun_cleanup);
3738 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3739 MODULE_AUTHOR(DRV_COPYRIGHT);
3740 MODULE_LICENSE("GPL");
3741 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3742 MODULE_ALIAS("devname:net/tun");